Skip to main content

dakera_client/
client.rs

1//! Dakera client implementation
2
3use reqwest::{Client, StatusCode};
4use std::sync::{Arc, Mutex};
5use std::time::Duration;
6use tracing::{debug, instrument};
7
8use serde::Deserialize;
9
10use crate::error::{ClientError, Result, ServerErrorCode};
11use crate::types::*;
12
13/// Default timeout for requests
14const DEFAULT_TIMEOUT_SECS: u64 = 30;
15
16/// Dakera client for interacting with the vector database
17#[derive(Debug, Clone)]
18pub struct DakeraClient {
19    /// HTTP client
20    pub(crate) client: Client,
21    /// Base URL of the Dakera server
22    pub(crate) base_url: String,
23    /// ODE-2: Base URL of the dakera-ode sidecar (optional)
24    pub(crate) ode_url: Option<String>,
25    /// Retry configuration (wired into API call sites in a follow-up; suppressed until then)
26    #[allow(dead_code)]
27    pub(crate) retry_config: RetryConfig,
28    /// OPS-1: last seen rate-limit headers (shared across clones)
29    pub(crate) last_rate_limit: Arc<Mutex<Option<RateLimitHeaders>>>,
30}
31
32impl DakeraClient {
33    /// Create a new client with the given base URL
34    ///
35    /// # Example
36    ///
37    /// ```rust,no_run
38    /// use dakera_client::DakeraClient;
39    ///
40    /// let client = DakeraClient::new("http://localhost:3000").unwrap();
41    /// ```
42    pub fn new(base_url: impl Into<String>) -> Result<Self> {
43        DakeraClientBuilder::new(base_url).build()
44    }
45
46    /// Create a new client builder for more configuration options
47    pub fn builder(base_url: impl Into<String>) -> DakeraClientBuilder {
48        DakeraClientBuilder::new(base_url)
49    }
50
51    // ========================================================================
52    // Health & Status
53    // ========================================================================
54
55    /// Check server health
56    #[instrument(skip(self))]
57    pub async fn health(&self) -> Result<HealthResponse> {
58        let url = format!("{}/health", self.base_url);
59        let response = self.client.get(&url).send().await?;
60
61        if response.status().is_success() {
62            Ok(response.json().await?)
63        } else {
64            // Health endpoint might return simple OK
65            Ok(HealthResponse {
66                healthy: true,
67                version: None,
68                uptime_seconds: None,
69            })
70        }
71    }
72
73    /// Check if server is ready
74    #[instrument(skip(self))]
75    pub async fn ready(&self) -> Result<ReadinessResponse> {
76        let url = format!("{}/health/ready", self.base_url);
77        let response = self.client.get(&url).send().await?;
78
79        if response.status().is_success() {
80            Ok(response.json().await?)
81        } else {
82            Ok(ReadinessResponse {
83                ready: false,
84                components: None,
85            })
86        }
87    }
88
89    /// Check if server is live
90    #[instrument(skip(self))]
91    pub async fn live(&self) -> Result<bool> {
92        let url = format!("{}/health/live", self.base_url);
93        let response = self.client.get(&url).send().await?;
94        Ok(response.status().is_success())
95    }
96
97    // ========================================================================
98    // Namespace Operations
99    // ========================================================================
100
101    /// List all namespaces
102    #[instrument(skip(self))]
103    pub async fn list_namespaces(&self) -> Result<Vec<String>> {
104        let url = format!("{}/v1/namespaces", self.base_url);
105        let response = self.client.get(&url).send().await?;
106        self.handle_response::<ListNamespacesResponse>(response)
107            .await
108            .map(|r| r.namespaces)
109    }
110
111    /// Get namespace information
112    #[instrument(skip(self))]
113    pub async fn get_namespace(&self, namespace: &str) -> Result<NamespaceInfo> {
114        let url = format!("{}/v1/namespaces/{}", self.base_url, namespace);
115        let response = self.client.get(&url).send().await?;
116        self.handle_response(response).await
117    }
118
119    /// Create a new namespace
120    #[instrument(skip(self, request))]
121    pub async fn create_namespace(
122        &self,
123        namespace: &str,
124        request: CreateNamespaceRequest,
125    ) -> Result<NamespaceInfo> {
126        let url = format!("{}/v1/namespaces/{}", self.base_url, namespace);
127        let response = self.client.post(&url).json(&request).send().await?;
128        self.handle_response(response).await
129    }
130
131    /// Create or update a namespace configuration (upsert semantics — v0.6.0).
132    ///
133    /// Creates the namespace if it does not exist, or updates its distance-metric
134    /// configuration if it already exists.  Dimension changes are rejected to
135    /// prevent silent data corruption.  Requires `Scope::Write`.
136    #[instrument(skip(self, request), fields(namespace = %namespace))]
137    pub async fn configure_namespace(
138        &self,
139        namespace: &str,
140        request: ConfigureNamespaceRequest,
141    ) -> Result<ConfigureNamespaceResponse> {
142        let url = format!("{}/v1/namespaces/{}", self.base_url, namespace);
143        let response = self.client.put(&url).json(&request).send().await?;
144        self.handle_response(response).await
145    }
146
147    // ========================================================================
148    // Vector Operations
149    // ========================================================================
150
151    /// Upsert vectors into a namespace
152    #[instrument(skip(self, request), fields(vector_count = request.vectors.len()))]
153    pub async fn upsert(&self, namespace: &str, request: UpsertRequest) -> Result<UpsertResponse> {
154        let url = format!("{}/v1/namespaces/{}/vectors", self.base_url, namespace);
155        debug!(
156            "Upserting {} vectors to {}",
157            request.vectors.len(),
158            namespace
159        );
160
161        let response = self.client.post(&url).json(&request).send().await?;
162        self.handle_response(response).await
163    }
164
165    /// Upsert a single vector (convenience method)
166    #[instrument(skip(self, vector))]
167    pub async fn upsert_one(&self, namespace: &str, vector: Vector) -> Result<UpsertResponse> {
168        self.upsert(namespace, UpsertRequest::single(vector)).await
169    }
170
171    /// Upsert vectors in column format (Turbopuffer-inspired)
172    ///
173    /// This format is more efficient for bulk upserts as it avoids repeating
174    /// field names for each vector. All arrays must have equal length.
175    ///
176    /// # Example
177    ///
178    /// ```rust,no_run
179    /// use dakera_client::{DakeraClient, ColumnUpsertRequest};
180    ///
181    /// # async fn example() -> Result<(), Box<dyn std::error::Error>> {
182    /// let client = DakeraClient::new("http://localhost:3000")?;
183    ///
184    /// let request = ColumnUpsertRequest::new(
185    ///     vec!["id1".to_string(), "id2".to_string(), "id3".to_string()],
186    ///     vec![
187    ///         vec![0.1, 0.2, 0.3],
188    ///         vec![0.4, 0.5, 0.6],
189    ///         vec![0.7, 0.8, 0.9],
190    ///     ],
191    /// )
192    /// .with_attribute("category", vec![
193    ///     serde_json::json!("A"),
194    ///     serde_json::json!("B"),
195    ///     serde_json::json!("A"),
196    /// ]);
197    ///
198    /// let response = client.upsert_columns("my-namespace", request).await?;
199    /// println!("Upserted {} vectors", response.upserted_count);
200    /// # Ok(())
201    /// # }
202    /// ```
203    #[instrument(skip(self, request), fields(namespace = %namespace, count = request.ids.len()))]
204    pub async fn upsert_columns(
205        &self,
206        namespace: &str,
207        request: ColumnUpsertRequest,
208    ) -> Result<UpsertResponse> {
209        let url = format!(
210            "{}/v1/namespaces/{}/upsert-columns",
211            self.base_url, namespace
212        );
213        debug!(
214            "Upserting {} vectors in column format to {}",
215            request.ids.len(),
216            namespace
217        );
218
219        let response = self.client.post(&url).json(&request).send().await?;
220        self.handle_response(response).await
221    }
222
223    /// Query for similar vectors
224    #[instrument(skip(self, request), fields(top_k = request.top_k))]
225    pub async fn query(&self, namespace: &str, request: QueryRequest) -> Result<QueryResponse> {
226        let url = format!("{}/v1/namespaces/{}/query", self.base_url, namespace);
227        debug!(
228            "Querying namespace {} for top {} results",
229            namespace, request.top_k
230        );
231
232        let response = self.client.post(&url).json(&request).send().await?;
233        self.handle_response(response).await
234    }
235
236    /// Simple query with just a vector and top_k (convenience method)
237    #[instrument(skip(self, vector))]
238    pub async fn query_simple(
239        &self,
240        namespace: &str,
241        vector: Vec<f32>,
242        top_k: u32,
243    ) -> Result<QueryResponse> {
244        self.query(namespace, QueryRequest::new(vector, top_k))
245            .await
246    }
247
248    /// Execute multiple queries in a single request
249    ///
250    /// This allows executing multiple vector similarity queries in parallel,
251    /// which is more efficient than making separate requests.
252    ///
253    /// # Example
254    ///
255    /// ```rust,no_run
256    /// use dakera_client::{DakeraClient, BatchQueryRequest, BatchQueryItem};
257    ///
258    /// # async fn example() -> Result<(), Box<dyn std::error::Error>> {
259    /// let client = DakeraClient::new("http://localhost:3000")?;
260    ///
261    /// let request = BatchQueryRequest::new(vec![
262    ///     BatchQueryItem::new(vec![0.1, 0.2, 0.3], 5).with_id("query1"),
263    ///     BatchQueryItem::new(vec![0.4, 0.5, 0.6], 10).with_id("query2"),
264    /// ]);
265    ///
266    /// let response = client.batch_query("my-namespace", request).await?;
267    /// println!("Executed {} queries in {}ms", response.query_count, response.total_latency_ms);
268    /// # Ok(())
269    /// # }
270    /// ```
271    #[instrument(skip(self, request), fields(namespace = %namespace, query_count = request.queries.len()))]
272    pub async fn batch_query(
273        &self,
274        namespace: &str,
275        request: BatchQueryRequest,
276    ) -> Result<BatchQueryResponse> {
277        let url = format!("{}/v1/namespaces/{}/batch-query", self.base_url, namespace);
278        debug!(
279            "Batch querying namespace {} with {} queries",
280            namespace,
281            request.queries.len()
282        );
283
284        let response = self.client.post(&url).json(&request).send().await?;
285        self.handle_response(response).await
286    }
287
288    /// Delete vectors by ID
289    #[instrument(skip(self, request), fields(id_count = request.ids.len()))]
290    pub async fn delete(&self, namespace: &str, request: DeleteRequest) -> Result<DeleteResponse> {
291        let url = format!(
292            "{}/v1/namespaces/{}/vectors/delete",
293            self.base_url, namespace
294        );
295        debug!("Deleting {} vectors from {}", request.ids.len(), namespace);
296
297        let response = self.client.post(&url).json(&request).send().await?;
298        self.handle_response(response).await
299    }
300
301    /// Delete a single vector by ID (convenience method)
302    #[instrument(skip(self))]
303    pub async fn delete_one(&self, namespace: &str, id: &str) -> Result<DeleteResponse> {
304        self.delete(namespace, DeleteRequest::single(id)).await
305    }
306
307    // ========================================================================
308    // Full-Text Search Operations
309    // ========================================================================
310
311    /// Index documents for full-text search
312    #[instrument(skip(self, request), fields(doc_count = request.documents.len()))]
313    pub async fn index_documents(
314        &self,
315        namespace: &str,
316        request: IndexDocumentsRequest,
317    ) -> Result<IndexDocumentsResponse> {
318        let url = format!(
319            "{}/v1/namespaces/{}/fulltext/index",
320            self.base_url, namespace
321        );
322        debug!(
323            "Indexing {} documents in {}",
324            request.documents.len(),
325            namespace
326        );
327
328        let response = self.client.post(&url).json(&request).send().await?;
329        self.handle_response(response).await
330    }
331
332    /// Index a single document (convenience method)
333    #[instrument(skip(self, document))]
334    pub async fn index_document(
335        &self,
336        namespace: &str,
337        document: Document,
338    ) -> Result<IndexDocumentsResponse> {
339        self.index_documents(
340            namespace,
341            IndexDocumentsRequest {
342                documents: vec![document],
343            },
344        )
345        .await
346    }
347
348    /// Perform full-text search
349    #[instrument(skip(self, request))]
350    pub async fn fulltext_search(
351        &self,
352        namespace: &str,
353        request: FullTextSearchRequest,
354    ) -> Result<FullTextSearchResponse> {
355        let url = format!(
356            "{}/v1/namespaces/{}/fulltext/search",
357            self.base_url, namespace
358        );
359        debug!("Full-text search in {} for: {}", namespace, request.query);
360
361        let response = self.client.post(&url).json(&request).send().await?;
362        self.handle_response(response).await
363    }
364
365    /// Simple full-text search (convenience method)
366    #[instrument(skip(self))]
367    pub async fn search_text(
368        &self,
369        namespace: &str,
370        query: &str,
371        top_k: u32,
372    ) -> Result<FullTextSearchResponse> {
373        self.fulltext_search(namespace, FullTextSearchRequest::new(query, top_k))
374            .await
375    }
376
377    /// Get full-text index statistics
378    #[instrument(skip(self))]
379    pub async fn fulltext_stats(&self, namespace: &str) -> Result<FullTextStats> {
380        let url = format!(
381            "{}/v1/namespaces/{}/fulltext/stats",
382            self.base_url, namespace
383        );
384        let response = self.client.get(&url).send().await?;
385        self.handle_response(response).await
386    }
387
388    /// Delete documents from full-text index
389    #[instrument(skip(self, request))]
390    pub async fn fulltext_delete(
391        &self,
392        namespace: &str,
393        request: DeleteRequest,
394    ) -> Result<DeleteResponse> {
395        let url = format!(
396            "{}/v1/namespaces/{}/fulltext/delete",
397            self.base_url, namespace
398        );
399        let response = self.client.post(&url).json(&request).send().await?;
400        self.handle_response(response).await
401    }
402
403    // ========================================================================
404    // Hybrid Search Operations
405    // ========================================================================
406
407    /// Perform hybrid search (vector + full-text)
408    #[instrument(skip(self, request), fields(top_k = request.top_k))]
409    pub async fn hybrid_search(
410        &self,
411        namespace: &str,
412        request: HybridSearchRequest,
413    ) -> Result<HybridSearchResponse> {
414        let url = format!("{}/v1/namespaces/{}/hybrid", self.base_url, namespace);
415        debug!(
416            "Hybrid search in {} with vector_weight={}",
417            namespace, request.vector_weight
418        );
419
420        let response = self.client.post(&url).json(&request).send().await?;
421        self.handle_response(response).await
422    }
423
424    // ========================================================================
425    // Multi-Vector Search Operations
426    // ========================================================================
427
428    /// Multi-vector search with positive/negative vectors and MMR
429    ///
430    /// This performs semantic search using multiple positive vectors (to search towards)
431    /// and optional negative vectors (to search away from). Supports MMR (Maximal Marginal
432    /// Relevance) for result diversity.
433    ///
434    /// # Example
435    ///
436    /// ```rust,no_run
437    /// use dakera_client::{DakeraClient, MultiVectorSearchRequest};
438    ///
439    /// # async fn example() -> Result<(), Box<dyn std::error::Error>> {
440    /// let client = DakeraClient::new("http://localhost:3000")?;
441    ///
442    /// // Search towards multiple concepts, away from others
443    /// let request = MultiVectorSearchRequest::new(vec![
444    ///     vec![0.1, 0.2, 0.3],  // positive vector 1
445    ///     vec![0.4, 0.5, 0.6],  // positive vector 2
446    /// ])
447    /// .with_negative_vectors(vec![
448    ///     vec![0.7, 0.8, 0.9],  // negative vector
449    /// ])
450    /// .with_top_k(10)
451    /// .with_mmr(0.7);  // Enable MMR with lambda=0.7
452    ///
453    /// let response = client.multi_vector_search("my-namespace", request).await?;
454    /// for result in response.results {
455    ///     println!("ID: {}, Score: {}", result.id, result.score);
456    /// }
457    /// # Ok(())
458    /// # }
459    /// ```
460    #[instrument(skip(self, request), fields(namespace = %namespace))]
461    pub async fn multi_vector_search(
462        &self,
463        namespace: &str,
464        request: MultiVectorSearchRequest,
465    ) -> Result<MultiVectorSearchResponse> {
466        let url = format!("{}/v1/namespaces/{}/multi-vector", self.base_url, namespace);
467        debug!(
468            "Multi-vector search in {} with {} positive vectors",
469            namespace,
470            request.positive_vectors.len()
471        );
472
473        let response = self.client.post(&url).json(&request).send().await?;
474        self.handle_response(response).await
475    }
476
477    // ========================================================================
478    // Aggregation Operations
479    // ========================================================================
480
481    /// Aggregate vectors with grouping (Turbopuffer-inspired)
482    ///
483    /// This performs aggregation queries on vector metadata, supporting
484    /// count, sum, avg, min, and max operations with optional grouping.
485    ///
486    /// # Example
487    ///
488    /// ```rust,no_run
489    /// use dakera_client::{DakeraClient, AggregationRequest};
490    ///
491    /// # async fn example() -> Result<(), Box<dyn std::error::Error>> {
492    /// let client = DakeraClient::new("http://localhost:3000")?;
493    ///
494    /// // Count all vectors and sum scores, grouped by category
495    /// let request = AggregationRequest::new()
496    ///     .with_count("total_count")
497    ///     .with_sum("total_score", "score")
498    ///     .with_avg("avg_score", "score")
499    ///     .with_group_by("category");
500    ///
501    /// let response = client.aggregate("my-namespace", request).await?;
502    /// if let Some(groups) = response.aggregation_groups {
503    ///     for group in groups {
504    ///         println!("Group: {:?}", group.group_key);
505    ///     }
506    /// }
507    /// # Ok(())
508    /// # }
509    /// ```
510    #[instrument(skip(self, request), fields(namespace = %namespace))]
511    pub async fn aggregate(
512        &self,
513        namespace: &str,
514        request: AggregationRequest,
515    ) -> Result<AggregationResponse> {
516        let url = format!("{}/v1/namespaces/{}/aggregate", self.base_url, namespace);
517        debug!(
518            "Aggregating in namespace {} with {} aggregations",
519            namespace,
520            request.aggregate_by.len()
521        );
522
523        let response = self.client.post(&url).json(&request).send().await?;
524        self.handle_response(response).await
525    }
526
527    // ========================================================================
528    // Unified Query Operations
529    // ========================================================================
530
531    /// Unified query with flexible ranking options (Turbopuffer-inspired)
532    ///
533    /// This provides a unified API for vector search (ANN/kNN), full-text search (BM25),
534    /// and attribute ordering. Supports combining multiple ranking functions with
535    /// Sum, Max, and Product operators.
536    ///
537    /// # Example
538    ///
539    /// ```rust,no_run
540    /// use dakera_client::{DakeraClient, UnifiedQueryRequest, SortDirection};
541    ///
542    /// # async fn example() -> Result<(), Box<dyn std::error::Error>> {
543    /// let client = DakeraClient::new("http://localhost:3000")?;
544    ///
545    /// // Vector ANN search
546    /// let request = UnifiedQueryRequest::vector_search(vec![0.1, 0.2, 0.3], 10);
547    /// let response = client.unified_query("my-namespace", request).await?;
548    ///
549    /// // Full-text BM25 search
550    /// let request = UnifiedQueryRequest::fulltext_search("content", "hello world", 10);
551    /// let response = client.unified_query("my-namespace", request).await?;
552    ///
553    /// // Attribute ordering with filter
554    /// let request = UnifiedQueryRequest::attribute_order("timestamp", SortDirection::Desc, 10)
555    ///     .with_filter(serde_json::json!({"category": {"$eq": "science"}}));
556    /// let response = client.unified_query("my-namespace", request).await?;
557    ///
558    /// for result in response.results {
559    ///     println!("ID: {}, Score: {:?}", result.id, result.dist);
560    /// }
561    /// # Ok(())
562    /// # }
563    /// ```
564    #[instrument(skip(self, request), fields(namespace = %namespace))]
565    pub async fn unified_query(
566        &self,
567        namespace: &str,
568        request: UnifiedQueryRequest,
569    ) -> Result<UnifiedQueryResponse> {
570        let url = format!(
571            "{}/v1/namespaces/{}/unified-query",
572            self.base_url, namespace
573        );
574        debug!(
575            "Unified query in namespace {} with top_k={}",
576            namespace, request.top_k
577        );
578
579        let response = self.client.post(&url).json(&request).send().await?;
580        self.handle_response(response).await
581    }
582
583    /// Simple vector search using the unified query API (convenience method)
584    ///
585    /// This is a shortcut for `unified_query` with a vector ANN search.
586    #[instrument(skip(self, vector))]
587    pub async fn unified_vector_search(
588        &self,
589        namespace: &str,
590        vector: Vec<f32>,
591        top_k: usize,
592    ) -> Result<UnifiedQueryResponse> {
593        self.unified_query(namespace, UnifiedQueryRequest::vector_search(vector, top_k))
594            .await
595    }
596
597    /// Simple full-text search using the unified query API (convenience method)
598    ///
599    /// This is a shortcut for `unified_query` with a BM25 full-text search.
600    #[instrument(skip(self))]
601    pub async fn unified_text_search(
602        &self,
603        namespace: &str,
604        field: &str,
605        query: &str,
606        top_k: usize,
607    ) -> Result<UnifiedQueryResponse> {
608        self.unified_query(
609            namespace,
610            UnifiedQueryRequest::fulltext_search(field, query, top_k),
611        )
612        .await
613    }
614
615    // ========================================================================
616    // Query Explain Operations
617    // ========================================================================
618
619    /// Explain query execution plan (similar to SQL EXPLAIN)
620    ///
621    /// This provides detailed information about how a query will be executed,
622    /// including index selection, execution stages, cost estimates, and
623    /// performance recommendations.
624    ///
625    /// # Example
626    ///
627    /// ```rust,no_run
628    /// use dakera_client::{DakeraClient, QueryExplainRequest};
629    ///
630    /// # async fn example() -> Result<(), Box<dyn std::error::Error>> {
631    /// let client = DakeraClient::new("http://localhost:3000")?;
632    ///
633    /// // Explain a vector search query
634    /// let request = QueryExplainRequest::vector_search(vec![0.1, 0.2, 0.3], 10)
635    ///     .with_verbose();
636    /// let plan = client.explain_query("my-namespace", request).await?;
637    ///
638    /// println!("Query plan: {}", plan.summary);
639    /// println!("Estimated time: {}ms", plan.cost_estimate.estimated_time_ms);
640    ///
641    /// for stage in &plan.stages {
642    ///     println!("Stage {}: {} - {}", stage.order, stage.name, stage.description);
643    /// }
644    ///
645    /// for rec in &plan.recommendations {
646    ///     println!("Recommendation ({}): {}", rec.priority, rec.description);
647    /// }
648    /// # Ok(())
649    /// # }
650    /// ```
651    #[instrument(skip(self, request), fields(namespace = %namespace))]
652    pub async fn explain_query(
653        &self,
654        namespace: &str,
655        request: QueryExplainRequest,
656    ) -> Result<QueryExplainResponse> {
657        let url = format!("{}/v1/namespaces/{}/explain", self.base_url, namespace);
658        debug!(
659            "Explaining query in namespace {} (query_type={:?}, top_k={})",
660            namespace, request.query_type, request.top_k
661        );
662
663        let response = self.client.post(&url).json(&request).send().await?;
664        self.handle_response(response).await
665    }
666
667    // ========================================================================
668    // Cache Warming Operations
669    // ========================================================================
670
671    /// Warm cache for vectors in a namespace
672    ///
673    /// This pre-loads vectors into cache tiers for faster subsequent access.
674    /// Supports priority levels and can run in the background.
675    ///
676    /// # Example
677    ///
678    /// ```rust,no_run
679    /// use dakera_client::{DakeraClient, WarmCacheRequest, WarmingPriority};
680    ///
681    /// # async fn example() -> Result<(), Box<dyn std::error::Error>> {
682    /// let client = DakeraClient::new("http://localhost:3000")?;
683    ///
684    /// // Warm entire namespace with high priority
685    /// let response = client.warm_cache(
686    ///     WarmCacheRequest::new("my-namespace")
687    ///         .with_priority(WarmingPriority::High)
688    /// ).await?;
689    ///
690    /// println!("Warmed {} entries", response.entries_warmed);
691    /// # Ok(())
692    /// # }
693    /// ```
694    #[instrument(skip(self, request), fields(namespace = %request.namespace, priority = ?request.priority))]
695    pub async fn warm_cache(&self, request: WarmCacheRequest) -> Result<WarmCacheResponse> {
696        let url = format!(
697            "{}/v1/namespaces/{}/cache/warm",
698            self.base_url, request.namespace
699        );
700        debug!(
701            "Warming cache for namespace {} with priority {:?}",
702            request.namespace, request.priority
703        );
704
705        let response = self.client.post(&url).json(&request).send().await?;
706        self.handle_response(response).await
707    }
708
709    /// Warm specific vectors by ID (convenience method)
710    #[instrument(skip(self, vector_ids))]
711    pub async fn warm_vectors(
712        &self,
713        namespace: &str,
714        vector_ids: Vec<String>,
715    ) -> Result<WarmCacheResponse> {
716        self.warm_cache(WarmCacheRequest::new(namespace).with_vector_ids(vector_ids))
717            .await
718    }
719
720    // ========================================================================
721    // Export Operations
722    // ========================================================================
723
724    /// Export vectors from a namespace with pagination
725    ///
726    /// This exports all vectors from a namespace, supporting pagination for
727    /// large datasets. Use the `next_cursor` from the response to fetch
728    /// subsequent pages.
729    ///
730    /// # Example
731    ///
732    /// ```rust,no_run
733    /// use dakera_client::{DakeraClient, ExportRequest};
734    ///
735    /// # async fn example() -> Result<(), Box<dyn std::error::Error>> {
736    /// let client = DakeraClient::new("http://localhost:3000")?;
737    ///
738    /// // Export first page of vectors
739    /// let mut request = ExportRequest::new().with_top_k(1000);
740    /// let response = client.export("my-namespace", request).await?;
741    ///
742    /// println!("Exported {} vectors", response.returned_count);
743    ///
744    /// // Fetch next page if available
745    /// if let Some(cursor) = response.next_cursor {
746    ///     let next_request = ExportRequest::new().with_cursor(cursor);
747    ///     let next_response = client.export("my-namespace", next_request).await?;
748    /// }
749    /// # Ok(())
750    /// # }
751    /// ```
752    #[instrument(skip(self, request), fields(namespace = %namespace))]
753    pub async fn export(&self, namespace: &str, request: ExportRequest) -> Result<ExportResponse> {
754        let url = format!("{}/v1/namespaces/{}/export", self.base_url, namespace);
755        debug!(
756            "Exporting vectors from namespace {} (top_k={}, cursor={:?})",
757            namespace, request.top_k, request.cursor
758        );
759
760        let response = self.client.post(&url).json(&request).send().await?;
761        self.handle_response(response).await
762    }
763
764    /// Export all vectors from a namespace (convenience method)
765    ///
766    /// This is a simple wrapper that exports with default settings.
767    #[instrument(skip(self))]
768    pub async fn export_all(&self, namespace: &str) -> Result<ExportResponse> {
769        self.export(namespace, ExportRequest::new()).await
770    }
771
772    // ========================================================================
773    // Operations
774    // ========================================================================
775
776    /// Get system diagnostics
777    #[instrument(skip(self))]
778    pub async fn diagnostics(&self) -> Result<SystemDiagnostics> {
779        let url = format!("{}/ops/diagnostics", self.base_url);
780        let response = self.client.get(&url).send().await?;
781        self.handle_response(response).await
782    }
783
784    /// List background jobs
785    #[instrument(skip(self))]
786    pub async fn list_jobs(&self) -> Result<Vec<JobInfo>> {
787        let url = format!("{}/ops/jobs", self.base_url);
788        let response = self.client.get(&url).send().await?;
789        self.handle_response(response).await
790    }
791
792    /// Get a specific job status
793    #[instrument(skip(self))]
794    pub async fn get_job(&self, job_id: &str) -> Result<Option<JobInfo>> {
795        let url = format!("{}/ops/jobs/{}", self.base_url, job_id);
796        let response = self.client.get(&url).send().await?;
797
798        if response.status() == StatusCode::NOT_FOUND {
799            return Ok(None);
800        }
801
802        self.handle_response(response).await.map(Some)
803    }
804
805    /// Trigger index compaction
806    #[instrument(skip(self, request))]
807    pub async fn compact(&self, request: CompactionRequest) -> Result<CompactionResponse> {
808        let url = format!("{}/ops/compact", self.base_url);
809        let response = self.client.post(&url).json(&request).send().await?;
810        self.handle_response(response).await
811    }
812
813    /// Request graceful shutdown
814    #[instrument(skip(self))]
815    pub async fn shutdown(&self) -> Result<()> {
816        let url = format!("{}/ops/shutdown", self.base_url);
817        let response = self.client.post(&url).send().await?;
818
819        if response.status().is_success() {
820            Ok(())
821        } else {
822            let status = response.status().as_u16();
823            let text = response.text().await.unwrap_or_default();
824            Err(ClientError::Server {
825                status,
826                message: text,
827                code: None,
828            })
829        }
830    }
831
832    // ========================================================================
833    // Fetch by ID
834    // ========================================================================
835
836    /// Fetch vectors by their IDs
837    #[instrument(skip(self, request), fields(id_count = request.ids.len()))]
838    pub async fn fetch(&self, namespace: &str, request: FetchRequest) -> Result<FetchResponse> {
839        let url = format!("{}/v1/namespaces/{}/fetch", self.base_url, namespace);
840        debug!("Fetching {} vectors from {}", request.ids.len(), namespace);
841        let response = self.client.post(&url).json(&request).send().await?;
842        self.handle_response(response).await
843    }
844
845    /// Fetch vectors by IDs (convenience method)
846    #[instrument(skip(self))]
847    pub async fn fetch_by_ids(&self, namespace: &str, ids: &[&str]) -> Result<Vec<Vector>> {
848        let request = FetchRequest::new(ids.iter().map(|s| s.to_string()).collect());
849        self.fetch(namespace, request).await.map(|r| r.vectors)
850    }
851
852    // ========================================================================
853    // Text Auto-Embedding Operations
854    // ========================================================================
855
856    /// Upsert text documents with automatic server-side embedding generation
857    #[instrument(skip(self, request), fields(doc_count = request.documents.len()))]
858    pub async fn upsert_text(
859        &self,
860        namespace: &str,
861        request: UpsertTextRequest,
862    ) -> Result<TextUpsertResponse> {
863        let url = format!("{}/v1/namespaces/{}/upsert-text", self.base_url, namespace);
864        debug!(
865            "Upserting {} text documents to {}",
866            request.documents.len(),
867            namespace
868        );
869        let response = self.client.post(&url).json(&request).send().await?;
870        self.handle_response(response).await
871    }
872
873    /// Query using natural language text with automatic server-side embedding
874    #[instrument(skip(self, request), fields(top_k = request.top_k))]
875    pub async fn query_text(
876        &self,
877        namespace: &str,
878        request: QueryTextRequest,
879    ) -> Result<TextQueryResponse> {
880        let url = format!("{}/v1/namespaces/{}/query-text", self.base_url, namespace);
881        debug!("Text query in {} for: {}", namespace, request.text);
882        let response = self.client.post(&url).json(&request).send().await?;
883        self.handle_response(response).await
884    }
885
886    /// Query text (convenience method)
887    #[instrument(skip(self))]
888    pub async fn query_text_simple(
889        &self,
890        namespace: &str,
891        text: &str,
892        top_k: u32,
893    ) -> Result<TextQueryResponse> {
894        self.query_text(namespace, QueryTextRequest::new(text, top_k))
895            .await
896    }
897
898    /// Execute multiple text queries with automatic embedding in a single request
899    #[instrument(skip(self, request), fields(query_count = request.queries.len()))]
900    pub async fn batch_query_text(
901        &self,
902        namespace: &str,
903        request: BatchQueryTextRequest,
904    ) -> Result<BatchQueryTextResponse> {
905        let url = format!(
906            "{}/v1/namespaces/{}/batch-query-text",
907            self.base_url, namespace
908        );
909        debug!(
910            "Batch text query in {} with {} queries",
911            namespace,
912            request.queries.len()
913        );
914        let response = self.client.post(&url).json(&request).send().await?;
915        self.handle_response(response).await
916    }
917
918    // ========================================================================
919    // CE-4: GLiNER Entity Extraction
920    // ========================================================================
921
922    /// Configure namespace-level entity extraction settings (CE-4).
923    ///
924    /// Sends `PATCH /v1/namespaces/{namespace}/config` with the provided
925    /// [`NamespaceNerConfig`].
926    #[instrument(skip(self, config))]
927    pub async fn configure_namespace_ner(
928        &self,
929        namespace: &str,
930        config: NamespaceNerConfig,
931    ) -> Result<serde_json::Value> {
932        let url = format!("{}/v1/namespaces/{}/config", self.base_url, namespace);
933        let response = self.client.patch(&url).json(&config).send().await?;
934        self.handle_response(response).await
935    }
936
937    /// Extract entities from arbitrary text using the GLiNER pipeline (CE-4).
938    ///
939    /// Sends `POST /v1/memories/extract` with the supplied text and optional
940    /// entity type list.
941    #[instrument(skip(self, text, entity_types))]
942    pub async fn extract_entities(
943        &self,
944        text: &str,
945        entity_types: Option<Vec<String>>,
946    ) -> Result<EntityExtractionResponse> {
947        let url = format!("{}/v1/memories/extract", self.base_url);
948        let body = serde_json::json!({
949            "text": text,
950            "entity_types": entity_types,
951        });
952        let response = self.client.post(&url).json(&body).send().await?;
953        self.handle_response(response).await
954    }
955
956    /// Retrieve entity tags associated with a stored memory (CE-4).
957    ///
958    /// Sends `GET /v1/memory/entities/{memory_id}`.
959    #[instrument(skip(self))]
960    pub async fn memory_entities(&self, memory_id: &str) -> Result<MemoryEntitiesResponse> {
961        let url = format!("{}/v1/memory/entities/{}", self.base_url, memory_id);
962        let response = self.client.get(&url).send().await?;
963        self.handle_response(response).await
964    }
965
966    // ========================================================================
967    // Private Helpers
968    // ========================================================================
969
970    /// Rate-limit headers from the most recent API response (OPS-1).
971    ///
972    /// Returns `None` until the first successful request has been made.
973    pub fn last_rate_limit_headers(&self) -> Option<RateLimitHeaders> {
974        self.last_rate_limit.lock().ok()?.clone()
975    }
976
977    /// Handle response and deserialize JSON
978    pub(crate) async fn handle_response<T: serde::de::DeserializeOwned>(
979        &self,
980        response: reqwest::Response,
981    ) -> Result<T> {
982        let status = response.status();
983
984        // OPS-1: capture rate-limit headers before consuming the response body
985        if let Ok(mut guard) = self.last_rate_limit.lock() {
986            *guard = Some(RateLimitHeaders::from_response(&response));
987        }
988
989        if status.is_success() {
990            Ok(response.json().await?)
991        } else {
992            let status_code = status.as_u16();
993            // Extract Retry-After before consuming response
994            let retry_after = response
995                .headers()
996                .get("Retry-After")
997                .and_then(|v| v.to_str().ok())
998                .and_then(|s| s.parse::<u64>().ok());
999            let text = response.text().await.unwrap_or_default();
1000
1001            if status_code == 429 {
1002                return Err(ClientError::RateLimitExceeded { retry_after });
1003            }
1004
1005            #[derive(Deserialize)]
1006            struct ErrorBody {
1007                error: Option<String>,
1008                code: Option<ServerErrorCode>,
1009            }
1010
1011            let (message, code) = if let Ok(body) = serde_json::from_str::<ErrorBody>(&text) {
1012                (body.error.unwrap_or_else(|| text.clone()), body.code)
1013            } else {
1014                (text, None)
1015            };
1016
1017            match status_code {
1018                401 => Err(ClientError::Server {
1019                    status: 401,
1020                    message,
1021                    code,
1022                }),
1023                403 => Err(ClientError::Authorization {
1024                    status: 403,
1025                    message,
1026                    code,
1027                }),
1028                404 => match &code {
1029                    Some(ServerErrorCode::NamespaceNotFound) => {
1030                        Err(ClientError::NamespaceNotFound(message))
1031                    }
1032                    Some(ServerErrorCode::VectorNotFound) => {
1033                        Err(ClientError::VectorNotFound(message))
1034                    }
1035                    _ => Err(ClientError::Server {
1036                        status: 404,
1037                        message,
1038                        code,
1039                    }),
1040                },
1041                _ => Err(ClientError::Server {
1042                    status: status_code,
1043                    message,
1044                    code,
1045                }),
1046            }
1047        }
1048    }
1049
1050    /// Handle response and return raw text body (for non-JSON endpoints like /v1/ops/metrics).
1051    pub(crate) async fn handle_text_response(&self, response: reqwest::Response) -> Result<String> {
1052        let status = response.status();
1053
1054        // OPS-1: capture rate-limit headers before consuming the response body
1055        if let Ok(mut guard) = self.last_rate_limit.lock() {
1056            *guard = Some(RateLimitHeaders::from_response(&response));
1057        }
1058
1059        let retry_after = response
1060            .headers()
1061            .get("Retry-After")
1062            .and_then(|v| v.to_str().ok())
1063            .and_then(|s| s.parse::<u64>().ok());
1064        let text = response.text().await.unwrap_or_default();
1065
1066        if status.is_success() {
1067            return Ok(text);
1068        }
1069
1070        let status_code = status.as_u16();
1071
1072        if status_code == 429 {
1073            return Err(ClientError::RateLimitExceeded { retry_after });
1074        }
1075
1076        #[derive(Deserialize)]
1077        struct ErrorBody {
1078            error: Option<String>,
1079            code: Option<ServerErrorCode>,
1080        }
1081
1082        let (message, code) = if let Ok(body) = serde_json::from_str::<ErrorBody>(&text) {
1083            (body.error.unwrap_or_else(|| text.clone()), body.code)
1084        } else {
1085            (text, None)
1086        };
1087
1088        match status_code {
1089            401 => Err(ClientError::Server {
1090                status: 401,
1091                message,
1092                code,
1093            }),
1094            403 => Err(ClientError::Authorization {
1095                status: 403,
1096                message,
1097                code,
1098            }),
1099            _ => Err(ClientError::Server {
1100                status: status_code,
1101                message,
1102                code,
1103            }),
1104        }
1105    }
1106
1107    /// Execute a fallible async operation with retry logic and exponential backoff.
1108    ///
1109    /// Retries on transient errors (5xx, rate-limit, connection/timeout).
1110    /// Respects the `Retry-After` header when the server returns HTTP 429.
1111    /// Does NOT retry on 4xx client errors (except 429).
1112    ///
1113    /// NOTE: API call-site wiring is deferred to a follow-up (infrastructure PR).
1114    #[allow(dead_code)]
1115    pub(crate) async fn execute_with_retry<F, Fut, T>(&self, f: F) -> Result<T>
1116    where
1117        F: Fn() -> Fut,
1118        Fut: std::future::Future<Output = Result<T>>,
1119    {
1120        let rc = &self.retry_config;
1121
1122        for attempt in 0..rc.max_retries {
1123            match f().await {
1124                Ok(v) => return Ok(v),
1125                Err(e) => {
1126                    let is_last = attempt == rc.max_retries - 1;
1127                    if is_last || !e.is_retryable() {
1128                        return Err(e);
1129                    }
1130
1131                    let wait = match &e {
1132                        ClientError::RateLimitExceeded {
1133                            retry_after: Some(secs),
1134                        } => Duration::from_secs(*secs),
1135                        _ => {
1136                            let base_ms = rc.base_delay.as_millis() as f64;
1137                            let backoff_ms = base_ms * 2f64.powi(attempt as i32);
1138                            let capped_ms = backoff_ms.min(rc.max_delay.as_millis() as f64);
1139                            let final_ms = if rc.jitter {
1140                                // Simple deterministic jitter: vary between 50% and 150%
1141                                let seed = (attempt as u64).wrapping_mul(6364136223846793005);
1142                                let factor = 0.5 + (seed % 1000) as f64 / 1000.0;
1143                                capped_ms * factor
1144                            } else {
1145                                capped_ms
1146                            };
1147                            Duration::from_millis(final_ms as u64)
1148                        }
1149                    };
1150
1151                    tokio::time::sleep(wait).await;
1152                }
1153            }
1154        }
1155
1156        // Unreachable: the loop always returns on the last attempt
1157        Err(ClientError::Config("retry loop exhausted".to_string()))
1158    }
1159}
1160
1161// ============================================================================
1162// ODE-2: GLiNER Entity Extraction (dakera-ode sidecar)
1163// ============================================================================
1164
1165impl DakeraClient {
1166    /// Extract named entities from text using the GLiNER sidecar (ODE-2).
1167    ///
1168    /// Calls `POST /ode/extract` on the dakera-ode sidecar. Requires
1169    /// [`ode_url`][DakeraClientBuilder::ode_url] to be set on the builder.
1170    ///
1171    /// Unlike the CE-4 server-side NER, this method calls the dedicated GLiNER
1172    /// sidecar and returns character offsets, model name, and processing time.
1173    ///
1174    /// # Errors
1175    ///
1176    /// Returns [`ClientError::Config`] if `ode_url` is not configured.
1177    pub async fn ode_extract_entities(
1178        &self,
1179        req: ExtractEntitiesRequest,
1180    ) -> Result<ExtractEntitiesResponse> {
1181        let ode_url = self.ode_url.as_deref().ok_or_else(|| {
1182            ClientError::Config(
1183                "ode_url must be configured to use extract_entities(). \
1184                 Call .ode_url(\"http://localhost:8080\") on the builder."
1185                    .to_string(),
1186            )
1187        })?;
1188        let url = format!("{}/ode/extract", ode_url);
1189        let response = self.client.post(&url).json(&req).send().await?;
1190        if response.status().is_success() {
1191            Ok(response.json::<ExtractEntitiesResponse>().await?)
1192        } else {
1193            let status = response.status().as_u16();
1194            let body = response.text().await.unwrap_or_default();
1195            Err(ClientError::Server {
1196                status,
1197                message: format!("ODE sidecar error: {}", body),
1198                code: None,
1199            })
1200        }
1201    }
1202}
1203
1204/// Builder for DakeraClient
1205#[derive(Debug)]
1206pub struct DakeraClientBuilder {
1207    base_url: String,
1208    ode_url: Option<String>,
1209    timeout: Duration,
1210    connect_timeout: Option<Duration>,
1211    retry_config: RetryConfig,
1212    user_agent: Option<String>,
1213}
1214
1215impl DakeraClientBuilder {
1216    /// Create a new builder
1217    pub fn new(base_url: impl Into<String>) -> Self {
1218        Self {
1219            base_url: base_url.into(),
1220            ode_url: None,
1221            timeout: Duration::from_secs(DEFAULT_TIMEOUT_SECS),
1222            connect_timeout: None,
1223            retry_config: RetryConfig::default(),
1224            user_agent: None,
1225        }
1226    }
1227
1228    /// Set the base URL of the dakera-ode sidecar (ODE-2).
1229    ///
1230    /// Required to call [`DakeraClient::extract_entities`].
1231    pub fn ode_url(mut self, ode_url: impl Into<String>) -> Self {
1232        self.ode_url = Some(ode_url.into().trim_end_matches('/').to_string());
1233        self
1234    }
1235
1236    /// Set the request timeout
1237    pub fn timeout(mut self, timeout: Duration) -> Self {
1238        self.timeout = timeout;
1239        self
1240    }
1241
1242    /// Set the request timeout in seconds
1243    pub fn timeout_secs(mut self, secs: u64) -> Self {
1244        self.timeout = Duration::from_secs(secs);
1245        self
1246    }
1247
1248    /// Set the connection establishment timeout (defaults to `timeout`).
1249    pub fn connect_timeout(mut self, timeout: Duration) -> Self {
1250        self.connect_timeout = Some(timeout);
1251        self
1252    }
1253
1254    /// Set fine-grained retry configuration.
1255    pub fn retry_config(mut self, config: RetryConfig) -> Self {
1256        self.retry_config = config;
1257        self
1258    }
1259
1260    /// Set the maximum number of retry attempts.
1261    pub fn max_retries(mut self, max_retries: u32) -> Self {
1262        self.retry_config.max_retries = max_retries;
1263        self
1264    }
1265
1266    /// Set a custom user agent
1267    pub fn user_agent(mut self, user_agent: impl Into<String>) -> Self {
1268        self.user_agent = Some(user_agent.into());
1269        self
1270    }
1271
1272    /// Build the client
1273    pub fn build(self) -> Result<DakeraClient> {
1274        // Normalize base URL (remove trailing slash)
1275        let base_url = self.base_url.trim_end_matches('/').to_string();
1276
1277        // Validate URL
1278        if !base_url.starts_with("http://") && !base_url.starts_with("https://") {
1279            return Err(ClientError::InvalidUrl(
1280                "URL must start with http:// or https://".to_string(),
1281            ));
1282        }
1283
1284        let user_agent = self
1285            .user_agent
1286            .unwrap_or_else(|| format!("dakera-client/{}", env!("CARGO_PKG_VERSION")));
1287
1288        let connect_timeout = self.connect_timeout.unwrap_or(self.timeout);
1289
1290        let client = Client::builder()
1291            .timeout(self.timeout)
1292            .connect_timeout(connect_timeout)
1293            .user_agent(user_agent)
1294            .build()
1295            .map_err(|e| ClientError::Config(e.to_string()))?;
1296
1297        Ok(DakeraClient {
1298            client,
1299            base_url,
1300            ode_url: self.ode_url,
1301            retry_config: self.retry_config,
1302            last_rate_limit: Arc::new(Mutex::new(None)),
1303        })
1304    }
1305}
1306
1307// ============================================================================
1308// SSE Streaming (CE-1)
1309// ============================================================================
1310
1311impl DakeraClient {
1312    /// Subscribe to namespace-scoped SSE events.
1313    ///
1314    /// Opens a long-lived connection to `GET /v1/namespaces/{namespace}/events`
1315    /// and returns a [`tokio::sync::mpsc::Receiver`] that yields
1316    /// [`DakeraEvent`] results as they arrive.  The background task exits when
1317    /// the server closes the stream or the receiver is dropped.
1318    ///
1319    /// Requires a Read-scoped API key.
1320    ///
1321    /// # Example
1322    ///
1323    /// ```rust,no_run
1324    /// use dakera_client::DakeraClient;
1325    ///
1326    /// #[tokio::main]
1327    /// async fn main() -> Result<(), Box<dyn std::error::Error>> {
1328    ///     let client = DakeraClient::new("http://localhost:3000")?;
1329    ///     let mut rx = client.stream_namespace_events("my-ns").await?;
1330    ///     while let Some(result) = rx.recv().await {
1331    ///         println!("{:?}", result?);
1332    ///     }
1333    ///     Ok(())
1334    /// }
1335    /// ```
1336    pub async fn stream_namespace_events(
1337        &self,
1338        namespace: &str,
1339    ) -> Result<tokio::sync::mpsc::Receiver<Result<crate::events::DakeraEvent>>> {
1340        let url = format!(
1341            "{}/v1/namespaces/{}/events",
1342            self.base_url,
1343            urlencoding::encode(namespace)
1344        );
1345        self.stream_sse(url).await
1346    }
1347
1348    /// Subscribe to the global SSE event stream (all namespaces).
1349    ///
1350    /// Opens a long-lived connection to `GET /ops/events` and returns a
1351    /// [`tokio::sync::mpsc::Receiver`] that yields [`DakeraEvent`] results.
1352    ///
1353    /// Requires an Admin-scoped API key.
1354    pub async fn stream_global_events(
1355        &self,
1356    ) -> Result<tokio::sync::mpsc::Receiver<Result<crate::events::DakeraEvent>>> {
1357        let url = format!("{}/ops/events", self.base_url);
1358        self.stream_sse(url).await
1359    }
1360
1361    /// Subscribe to the memory lifecycle SSE event stream (DASH-B).
1362    ///
1363    /// Opens a long-lived connection to `GET /v1/events/stream` and returns a
1364    /// [`tokio::sync::mpsc::Receiver`] that yields [`MemoryEvent`] results as
1365    /// they arrive.  The background task exits when the server closes the stream
1366    /// or the receiver is dropped.
1367    ///
1368    /// Requires a Read-scoped API key.
1369    pub async fn stream_memory_events(
1370        &self,
1371    ) -> Result<tokio::sync::mpsc::Receiver<Result<crate::events::MemoryEvent>>> {
1372        let url = format!("{}/v1/events/stream", self.base_url);
1373        self.stream_sse(url).await
1374    }
1375
1376    /// Low-level generic SSE streaming helper.
1377    pub(crate) async fn stream_sse<T>(
1378        &self,
1379        url: String,
1380    ) -> Result<tokio::sync::mpsc::Receiver<Result<T>>>
1381    where
1382        T: serde::de::DeserializeOwned + Send + 'static,
1383    {
1384        use futures_util::StreamExt;
1385
1386        let response = self
1387            .client
1388            .get(&url)
1389            .header("Accept", "text/event-stream")
1390            .header("Cache-Control", "no-cache")
1391            .send()
1392            .await?;
1393
1394        if !response.status().is_success() {
1395            let status = response.status().as_u16();
1396            let body = response.text().await.unwrap_or_default();
1397            return Err(ClientError::Server {
1398                status,
1399                message: body,
1400                code: None,
1401            });
1402        }
1403
1404        let (tx, rx) = tokio::sync::mpsc::channel(64);
1405
1406        tokio::spawn(async move {
1407            let mut byte_stream = response.bytes_stream();
1408            let mut remaining = String::new();
1409            let mut data_lines: Vec<String> = Vec::new();
1410
1411            while let Some(chunk) = byte_stream.next().await {
1412                match chunk {
1413                    Ok(bytes) => {
1414                        remaining.push_str(&String::from_utf8_lossy(&bytes));
1415                        while let Some(pos) = remaining.find('\n') {
1416                            let raw = &remaining[..pos];
1417                            let line = raw.trim_end_matches('\r').to_string();
1418                            remaining = remaining[pos + 1..].to_string();
1419
1420                            if line.starts_with(':') {
1421                                // SSE comment / heartbeat — skip
1422                            } else if let Some(data) = line.strip_prefix("data:") {
1423                                data_lines.push(data.trim_start().to_string());
1424                            } else if line.is_empty() {
1425                                if !data_lines.is_empty() {
1426                                    let payload = data_lines.join("\n");
1427                                    data_lines.clear();
1428                                    let result = serde_json::from_str::<T>(&payload)
1429                                        .map_err(ClientError::Json);
1430                                    if tx.send(result).await.is_err() {
1431                                        return; // receiver dropped
1432                                    }
1433                                }
1434                            } else {
1435                                // Unrecognised field (e.g. "event:") — ignore
1436                            }
1437                        }
1438                    }
1439                    Err(e) => {
1440                        let _ = tx.send(Err(ClientError::Http(e))).await;
1441                        return;
1442                    }
1443                }
1444            }
1445        });
1446
1447        Ok(rx)
1448    }
1449}
1450
1451#[cfg(test)]
1452mod tests {
1453    use super::*;
1454
1455    #[test]
1456    fn test_client_builder() {
1457        let client = DakeraClient::new("http://localhost:3000");
1458        assert!(client.is_ok());
1459    }
1460
1461    #[test]
1462    fn test_client_builder_with_options() {
1463        let client = DakeraClient::builder("http://localhost:3000")
1464            .timeout_secs(60)
1465            .user_agent("test-client/1.0")
1466            .build();
1467        assert!(client.is_ok());
1468    }
1469
1470    #[test]
1471    fn test_client_builder_invalid_url() {
1472        let client = DakeraClient::new("invalid-url");
1473        assert!(client.is_err());
1474    }
1475
1476    #[test]
1477    fn test_client_builder_trailing_slash() {
1478        let client = DakeraClient::new("http://localhost:3000/").unwrap();
1479        assert!(!client.base_url.ends_with('/'));
1480    }
1481
1482    #[test]
1483    fn test_vector_creation() {
1484        let v = Vector::new("test", vec![0.1, 0.2, 0.3]);
1485        assert_eq!(v.id, "test");
1486        assert_eq!(v.values.len(), 3);
1487        assert!(v.metadata.is_none());
1488    }
1489
1490    #[test]
1491    fn test_query_request_builder() {
1492        let req = QueryRequest::new(vec![0.1, 0.2], 10)
1493            .with_filter(serde_json::json!({"category": "test"}))
1494            .include_metadata(false);
1495
1496        assert_eq!(req.top_k, 10);
1497        assert!(req.filter.is_some());
1498        assert!(!req.include_metadata);
1499    }
1500
1501    #[test]
1502    fn test_hybrid_search_request() {
1503        let req = HybridSearchRequest::new(vec![0.1], "test query", 5).with_vector_weight(0.7);
1504
1505        assert_eq!(req.vector_weight, 0.7);
1506        assert_eq!(req.text, "test query");
1507        assert!(req.vector.is_some());
1508    }
1509
1510    #[test]
1511    fn test_hybrid_search_weight_clamping() {
1512        let req = HybridSearchRequest::new(vec![0.1], "test", 5).with_vector_weight(1.5); // Should be clamped to 1.0
1513
1514        assert_eq!(req.vector_weight, 1.0);
1515    }
1516
1517    #[test]
1518    fn test_hybrid_search_text_only() {
1519        let req = HybridSearchRequest::text_only("bm25 query", 10);
1520
1521        assert!(req.vector.is_none());
1522        assert_eq!(req.text, "bm25 query");
1523        assert_eq!(req.top_k, 10);
1524        // Verify vector is not serialised
1525        let json = serde_json::to_value(&req).unwrap();
1526        assert!(json.get("vector").is_none());
1527    }
1528
1529    #[test]
1530    fn test_text_document_builder() {
1531        let doc = TextDocument::new("doc1", "Hello world").with_ttl(3600);
1532
1533        assert_eq!(doc.id, "doc1");
1534        assert_eq!(doc.text, "Hello world");
1535        assert_eq!(doc.ttl_seconds, Some(3600));
1536        assert!(doc.metadata.is_none());
1537    }
1538
1539    #[test]
1540    fn test_upsert_text_request_builder() {
1541        let docs = vec![
1542            TextDocument::new("doc1", "Hello"),
1543            TextDocument::new("doc2", "World"),
1544        ];
1545        let req = UpsertTextRequest::new(docs).with_model(EmbeddingModel::BgeSmall);
1546
1547        assert_eq!(req.documents.len(), 2);
1548        assert_eq!(req.model, Some(EmbeddingModel::BgeSmall));
1549    }
1550
1551    #[test]
1552    fn test_query_text_request_builder() {
1553        let req = QueryTextRequest::new("semantic search query", 5)
1554            .with_filter(serde_json::json!({"category": "docs"}))
1555            .include_vectors(true)
1556            .with_model(EmbeddingModel::E5Small);
1557
1558        assert_eq!(req.text, "semantic search query");
1559        assert_eq!(req.top_k, 5);
1560        assert!(req.filter.is_some());
1561        assert!(req.include_vectors);
1562        assert_eq!(req.model, Some(EmbeddingModel::E5Small));
1563    }
1564
1565    #[test]
1566    fn test_fetch_request_builder() {
1567        let req = FetchRequest::new(vec!["id1".to_string(), "id2".to_string()]);
1568
1569        assert_eq!(req.ids.len(), 2);
1570        assert!(req.include_values);
1571        assert!(req.include_metadata);
1572    }
1573
1574    #[test]
1575    fn test_create_namespace_request_builder() {
1576        let req = CreateNamespaceRequest::new()
1577            .with_dimensions(384)
1578            .with_index_type("hnsw");
1579
1580        assert_eq!(req.dimensions, Some(384));
1581        assert_eq!(req.index_type.as_deref(), Some("hnsw"));
1582    }
1583
1584    #[test]
1585    fn test_batch_query_text_request() {
1586        let req =
1587            BatchQueryTextRequest::new(vec!["query one".to_string(), "query two".to_string()], 10);
1588
1589        assert_eq!(req.queries.len(), 2);
1590        assert_eq!(req.top_k, 10);
1591        assert!(!req.include_vectors);
1592        assert!(req.model.is_none());
1593    }
1594
1595    // =========================================================================
1596    // RetryConfig tests
1597    // =========================================================================
1598
1599    #[test]
1600    fn test_retry_config_defaults() {
1601        let rc = RetryConfig::default();
1602        assert_eq!(rc.max_retries, 3);
1603        assert_eq!(rc.base_delay, Duration::from_millis(100));
1604        assert_eq!(rc.max_delay, Duration::from_secs(60));
1605        assert!(rc.jitter);
1606    }
1607
1608    #[test]
1609    fn test_builder_connect_timeout() {
1610        let client = DakeraClient::builder("http://localhost:3000")
1611            .connect_timeout(Duration::from_secs(5))
1612            .timeout_secs(30)
1613            .build()
1614            .unwrap();
1615        // Client was built successfully with separate connect timeout
1616        assert!(client.base_url.starts_with("http"));
1617    }
1618
1619    #[test]
1620    fn test_builder_max_retries() {
1621        let client = DakeraClient::builder("http://localhost:3000")
1622            .max_retries(5)
1623            .build()
1624            .unwrap();
1625        assert_eq!(client.retry_config.max_retries, 5);
1626    }
1627
1628    #[test]
1629    fn test_builder_retry_config() {
1630        let rc = RetryConfig {
1631            max_retries: 7,
1632            base_delay: Duration::from_millis(200),
1633            max_delay: Duration::from_secs(30),
1634            jitter: false,
1635        };
1636        let client = DakeraClient::builder("http://localhost:3000")
1637            .retry_config(rc)
1638            .build()
1639            .unwrap();
1640        assert_eq!(client.retry_config.max_retries, 7);
1641        assert!(!client.retry_config.jitter);
1642    }
1643
1644    #[test]
1645    fn test_rate_limit_error_retryable() {
1646        let e = ClientError::RateLimitExceeded { retry_after: None };
1647        assert!(e.is_retryable());
1648    }
1649
1650    #[test]
1651    fn test_rate_limit_error_with_retry_after_zero() {
1652        // retry_after: Some(0) should still be Some, not treated as missing
1653        let e = ClientError::RateLimitExceeded {
1654            retry_after: Some(0),
1655        };
1656        assert!(e.is_retryable());
1657        if let ClientError::RateLimitExceeded {
1658            retry_after: Some(secs),
1659        } = &e
1660        {
1661            assert_eq!(*secs, 0u64);
1662        } else {
1663            panic!("unexpected variant");
1664        }
1665    }
1666
1667    #[tokio::test]
1668    async fn test_execute_with_retry_succeeds_immediately() {
1669        let client = DakeraClient::builder("http://localhost:3000")
1670            .max_retries(3)
1671            .build()
1672            .unwrap();
1673
1674        let call_count = std::sync::Arc::new(std::sync::atomic::AtomicU32::new(0));
1675        let cc = call_count.clone();
1676        let result = client
1677            .execute_with_retry(|| {
1678                let cc = cc.clone();
1679                async move {
1680                    cc.fetch_add(1, std::sync::atomic::Ordering::SeqCst);
1681                    Ok::<u32, ClientError>(42)
1682                }
1683            })
1684            .await;
1685        assert_eq!(result.unwrap(), 42);
1686        assert_eq!(call_count.load(std::sync::atomic::Ordering::SeqCst), 1);
1687    }
1688
1689    #[tokio::test]
1690    async fn test_execute_with_retry_no_retry_on_4xx() {
1691        let client = DakeraClient::builder("http://localhost:3000")
1692            .max_retries(3)
1693            .build()
1694            .unwrap();
1695
1696        let call_count = std::sync::Arc::new(std::sync::atomic::AtomicU32::new(0));
1697        let cc = call_count.clone();
1698        let result = client
1699            .execute_with_retry(|| {
1700                let cc = cc.clone();
1701                async move {
1702                    cc.fetch_add(1, std::sync::atomic::Ordering::SeqCst);
1703                    Err::<u32, ClientError>(ClientError::Server {
1704                        status: 400,
1705                        message: "bad request".to_string(),
1706                        code: None,
1707                    })
1708                }
1709            })
1710            .await;
1711        assert!(result.is_err());
1712        // Should not retry on 4xx
1713        assert_eq!(call_count.load(std::sync::atomic::Ordering::SeqCst), 1);
1714    }
1715
1716    #[tokio::test]
1717    async fn test_execute_with_retry_retries_on_5xx() {
1718        let client = DakeraClient::builder("http://localhost:3000")
1719            .retry_config(RetryConfig {
1720                max_retries: 3,
1721                base_delay: Duration::from_millis(0),
1722                max_delay: Duration::from_millis(0),
1723                jitter: false,
1724            })
1725            .build()
1726            .unwrap();
1727
1728        let call_count = std::sync::Arc::new(std::sync::atomic::AtomicU32::new(0));
1729        let cc = call_count.clone();
1730        let result = client
1731            .execute_with_retry(|| {
1732                let cc = cc.clone();
1733                async move {
1734                    let n = cc.fetch_add(1, std::sync::atomic::Ordering::SeqCst);
1735                    if n < 2 {
1736                        Err::<u32, ClientError>(ClientError::Server {
1737                            status: 503,
1738                            message: "unavailable".to_string(),
1739                            code: None,
1740                        })
1741                    } else {
1742                        Ok(99)
1743                    }
1744                }
1745            })
1746            .await;
1747        assert_eq!(result.unwrap(), 99);
1748        assert_eq!(call_count.load(std::sync::atomic::Ordering::SeqCst), 3);
1749    }
1750
1751    // =========================================================================
1752    // CE-2: Batch Recall / Forget (v0.7.0)
1753    // =========================================================================
1754
1755    #[test]
1756    fn test_batch_recall_request_new() {
1757        use crate::memory::BatchRecallRequest;
1758        let req = BatchRecallRequest::new("agent-1");
1759        assert_eq!(req.agent_id, "agent-1");
1760        assert_eq!(req.limit, 100);
1761    }
1762
1763    #[test]
1764    fn test_batch_recall_request_builder() {
1765        use crate::memory::{BatchMemoryFilter, BatchRecallRequest};
1766        let filter = BatchMemoryFilter::default()
1767            .with_tags(vec!["qa".to_string()])
1768            .with_min_importance(0.7);
1769        let req = BatchRecallRequest::new("agent-1")
1770            .with_filter(filter)
1771            .with_limit(50);
1772        assert_eq!(req.agent_id, "agent-1");
1773        assert_eq!(req.limit, 50);
1774        assert_eq!(
1775            req.filter.tags.as_deref(),
1776            Some(["qa".to_string()].as_slice())
1777        );
1778        assert_eq!(req.filter.min_importance, Some(0.7));
1779    }
1780
1781    #[test]
1782    fn test_batch_recall_request_serialization() {
1783        use crate::memory::{BatchMemoryFilter, BatchRecallRequest};
1784        let filter = BatchMemoryFilter::default().with_min_importance(0.5);
1785        let req = BatchRecallRequest::new("agent-1")
1786            .with_filter(filter)
1787            .with_limit(25);
1788        let json = serde_json::to_value(&req).unwrap();
1789        assert_eq!(json["agent_id"], "agent-1");
1790        assert_eq!(json["limit"], 25);
1791        assert_eq!(json["filter"]["min_importance"], 0.5);
1792    }
1793
1794    #[test]
1795    fn test_batch_forget_request_new() {
1796        use crate::memory::{BatchForgetRequest, BatchMemoryFilter};
1797        let filter = BatchMemoryFilter::default().with_min_importance(0.1);
1798        let req = BatchForgetRequest::new("agent-1", filter);
1799        assert_eq!(req.agent_id, "agent-1");
1800        assert_eq!(req.filter.min_importance, Some(0.1));
1801    }
1802
1803    #[test]
1804    fn test_batch_forget_request_serialization() {
1805        use crate::memory::{BatchForgetRequest, BatchMemoryFilter};
1806        let filter = BatchMemoryFilter {
1807            created_before: Some(1_700_000_000),
1808            ..Default::default()
1809        };
1810        let req = BatchForgetRequest::new("agent-1", filter);
1811        let json = serde_json::to_value(&req).unwrap();
1812        assert_eq!(json["agent_id"], "agent-1");
1813        assert_eq!(json["filter"]["created_before"], 1_700_000_000u64);
1814    }
1815
1816    #[test]
1817    fn test_batch_recall_response_deserialization() {
1818        use crate::memory::BatchRecallResponse;
1819        let json = serde_json::json!({
1820            "memories": [],
1821            "total": 42,
1822            "filtered": 7
1823        });
1824        let resp: BatchRecallResponse = serde_json::from_value(json).unwrap();
1825        assert_eq!(resp.total, 42);
1826        assert_eq!(resp.filtered, 7);
1827        assert!(resp.memories.is_empty());
1828    }
1829
1830    #[test]
1831    fn test_batch_forget_response_deserialization() {
1832        use crate::memory::BatchForgetResponse;
1833        let json = serde_json::json!({ "deleted_count": 13 });
1834        let resp: BatchForgetResponse = serde_json::from_value(json).unwrap();
1835        assert_eq!(resp.deleted_count, 13);
1836    }
1837
1838    // =========================================================================
1839    // OPS-1: RateLimitHeaders (v0.7.0)
1840    // =========================================================================
1841
1842    #[test]
1843    fn test_rate_limit_headers_default_all_none() {
1844        use crate::types::RateLimitHeaders;
1845        let rl = RateLimitHeaders {
1846            limit: None,
1847            remaining: None,
1848            reset: None,
1849            quota_used: None,
1850            quota_limit: None,
1851        };
1852        assert!(rl.limit.is_none());
1853        assert!(rl.remaining.is_none());
1854        assert!(rl.reset.is_none());
1855        assert!(rl.quota_used.is_none());
1856        assert!(rl.quota_limit.is_none());
1857    }
1858
1859    #[test]
1860    fn test_rate_limit_headers_populated() {
1861        use crate::types::RateLimitHeaders;
1862        let rl = RateLimitHeaders {
1863            limit: Some(1000),
1864            remaining: Some(750),
1865            reset: Some(1_700_000_060),
1866            quota_used: Some(500),
1867            quota_limit: Some(10_000),
1868        };
1869        assert_eq!(rl.limit, Some(1000));
1870        assert_eq!(rl.remaining, Some(750));
1871        assert_eq!(rl.reset, Some(1_700_000_060));
1872        assert_eq!(rl.quota_used, Some(500));
1873        assert_eq!(rl.quota_limit, Some(10_000));
1874    }
1875
1876    #[test]
1877    fn test_last_rate_limit_headers_initially_none() {
1878        let client = DakeraClient::new("http://localhost:3000").unwrap();
1879        assert!(client.last_rate_limit_headers().is_none());
1880    }
1881
1882    // =========================================================================
1883    // CE-4: GLiNER Entity Extraction
1884    // =========================================================================
1885
1886    #[test]
1887    fn test_namespace_ner_config_default() {
1888        use crate::types::NamespaceNerConfig;
1889        let cfg = NamespaceNerConfig::default();
1890        assert!(!cfg.extract_entities);
1891        assert!(cfg.entity_types.is_none());
1892    }
1893
1894    #[test]
1895    fn test_namespace_ner_config_serialization_skip_none() {
1896        use crate::types::NamespaceNerConfig;
1897        let cfg = NamespaceNerConfig {
1898            extract_entities: true,
1899            entity_types: None,
1900        };
1901        let json = serde_json::to_value(&cfg).unwrap();
1902        assert_eq!(json["extract_entities"], true);
1903        // entity_types should be omitted when None
1904        assert!(json.get("entity_types").is_none());
1905    }
1906
1907    #[test]
1908    fn test_namespace_ner_config_serialization_with_types() {
1909        use crate::types::NamespaceNerConfig;
1910        let cfg = NamespaceNerConfig {
1911            extract_entities: true,
1912            entity_types: Some(vec!["PERSON".to_string(), "ORG".to_string()]),
1913        };
1914        let json = serde_json::to_value(&cfg).unwrap();
1915        assert_eq!(json["extract_entities"], true);
1916        assert_eq!(json["entity_types"][0], "PERSON");
1917        assert_eq!(json["entity_types"][1], "ORG");
1918    }
1919
1920    #[test]
1921    fn test_extracted_entity_deserialization() {
1922        use crate::types::ExtractedEntity;
1923        let json = serde_json::json!({
1924            "entity_type": "PERSON",
1925            "value": "Alice",
1926            "score": 0.95
1927        });
1928        let entity: ExtractedEntity = serde_json::from_value(json).unwrap();
1929        assert_eq!(entity.entity_type, "PERSON");
1930        assert_eq!(entity.value, "Alice");
1931        assert!((entity.score - 0.95).abs() < f64::EPSILON);
1932    }
1933
1934    #[test]
1935    fn test_entity_extraction_response_deserialization() {
1936        use crate::types::EntityExtractionResponse;
1937        let json = serde_json::json!({
1938            "entities": [
1939                { "entity_type": "PERSON", "value": "Bob", "score": 0.9 },
1940                { "entity_type": "ORG",    "value": "Acme", "score": 0.87 }
1941            ]
1942        });
1943        let resp: EntityExtractionResponse = serde_json::from_value(json).unwrap();
1944        assert_eq!(resp.entities.len(), 2);
1945        assert_eq!(resp.entities[0].entity_type, "PERSON");
1946        assert_eq!(resp.entities[1].value, "Acme");
1947    }
1948
1949    #[test]
1950    fn test_memory_entities_response_deserialization() {
1951        use crate::types::MemoryEntitiesResponse;
1952        let json = serde_json::json!({
1953            "memory_id": "mem-abc-123",
1954            "entities": [
1955                { "entity_type": "LOC", "value": "London", "score": 0.88 }
1956            ]
1957        });
1958        let resp: MemoryEntitiesResponse = serde_json::from_value(json).unwrap();
1959        assert_eq!(resp.memory_id, "mem-abc-123");
1960        assert_eq!(resp.entities.len(), 1);
1961        assert_eq!(resp.entities[0].entity_type, "LOC");
1962        assert_eq!(resp.entities[0].value, "London");
1963    }
1964
1965    #[test]
1966    fn test_configure_namespace_ner_url_pattern() {
1967        // Verify the client is constructable and base_url is correct
1968        let client = DakeraClient::new("http://localhost:3000").unwrap();
1969        let expected = "http://localhost:3000/v1/namespaces/my-ns/config";
1970        let actual = format!("{}/v1/namespaces/{}/config", client.base_url, "my-ns");
1971        assert_eq!(actual, expected);
1972    }
1973
1974    #[test]
1975    fn test_extract_entities_url_pattern() {
1976        let client = DakeraClient::new("http://localhost:3000").unwrap();
1977        let expected = "http://localhost:3000/v1/memories/extract";
1978        let actual = format!("{}/v1/memories/extract", client.base_url);
1979        assert_eq!(actual, expected);
1980    }
1981
1982    #[test]
1983    fn test_memory_entities_url_pattern() {
1984        let client = DakeraClient::new("http://localhost:3000").unwrap();
1985        let memory_id = "mem-xyz-789";
1986        let expected = "http://localhost:3000/v1/memory/entities/mem-xyz-789";
1987        let actual = format!("{}/v1/memory/entities/{}", client.base_url, memory_id);
1988        assert_eq!(actual, expected);
1989    }
1990
1991    // ========================================================================
1992    // INT-1 Memory Feedback Loop tests
1993    // ========================================================================
1994
1995    #[test]
1996    fn test_feedback_signal_serialization() {
1997        use crate::types::FeedbackSignal;
1998        let upvote = serde_json::to_value(FeedbackSignal::Upvote).unwrap();
1999        assert_eq!(upvote, serde_json::json!("upvote"));
2000        let downvote = serde_json::to_value(FeedbackSignal::Downvote).unwrap();
2001        assert_eq!(downvote, serde_json::json!("downvote"));
2002        let flag = serde_json::to_value(FeedbackSignal::Flag).unwrap();
2003        assert_eq!(flag, serde_json::json!("flag"));
2004    }
2005
2006    #[test]
2007    fn test_feedback_signal_deserialization() {
2008        use crate::types::FeedbackSignal;
2009        let signal: FeedbackSignal = serde_json::from_str("\"upvote\"").unwrap();
2010        assert_eq!(signal, FeedbackSignal::Upvote);
2011        let signal: FeedbackSignal = serde_json::from_str("\"positive\"").unwrap();
2012        assert_eq!(signal, FeedbackSignal::Positive);
2013    }
2014
2015    #[test]
2016    fn test_feedback_response_deserialization() {
2017        use crate::types::{FeedbackResponse, FeedbackSignal};
2018        let json = serde_json::json!({
2019            "memory_id": "mem-abc",
2020            "new_importance": 0.92,
2021            "signal": "upvote"
2022        });
2023        let resp: FeedbackResponse = serde_json::from_value(json).unwrap();
2024        assert_eq!(resp.memory_id, "mem-abc");
2025        assert!((resp.new_importance - 0.92).abs() < f32::EPSILON);
2026        assert_eq!(resp.signal, FeedbackSignal::Upvote);
2027    }
2028
2029    #[test]
2030    fn test_feedback_history_response_deserialization() {
2031        use crate::types::{FeedbackHistoryResponse, FeedbackSignal};
2032        let json = serde_json::json!({
2033            "memory_id": "mem-abc",
2034            "entries": [
2035                {"signal": "upvote", "timestamp": 1774000000_u64, "old_importance": 0.5, "new_importance": 0.575},
2036                {"signal": "downvote", "timestamp": 1774001000_u64, "old_importance": 0.575, "new_importance": 0.489}
2037            ]
2038        });
2039        let resp: FeedbackHistoryResponse = serde_json::from_value(json).unwrap();
2040        assert_eq!(resp.memory_id, "mem-abc");
2041        assert_eq!(resp.entries.len(), 2);
2042        assert_eq!(resp.entries[0].signal, FeedbackSignal::Upvote);
2043        assert_eq!(resp.entries[1].signal, FeedbackSignal::Downvote);
2044    }
2045
2046    #[test]
2047    fn test_agent_feedback_summary_deserialization() {
2048        use crate::types::AgentFeedbackSummary;
2049        let json = serde_json::json!({
2050            "agent_id": "agent-1",
2051            "upvotes": 42_u64,
2052            "downvotes": 7_u64,
2053            "flags": 2_u64,
2054            "total_feedback": 51_u64,
2055            "health_score": 0.78
2056        });
2057        let summary: AgentFeedbackSummary = serde_json::from_value(json).unwrap();
2058        assert_eq!(summary.agent_id, "agent-1");
2059        assert_eq!(summary.upvotes, 42);
2060        assert_eq!(summary.total_feedback, 51);
2061        assert!((summary.health_score - 0.78).abs() < f32::EPSILON);
2062    }
2063
2064    #[test]
2065    fn test_feedback_health_response_deserialization() {
2066        use crate::types::FeedbackHealthResponse;
2067        let json = serde_json::json!({
2068            "agent_id": "agent-1",
2069            "health_score": 0.78,
2070            "memory_count": 120_usize,
2071            "avg_importance": 0.72
2072        });
2073        let health: FeedbackHealthResponse = serde_json::from_value(json).unwrap();
2074        assert_eq!(health.agent_id, "agent-1");
2075        assert!((health.health_score - 0.78).abs() < f32::EPSILON);
2076        assert_eq!(health.memory_count, 120);
2077    }
2078
2079    #[test]
2080    fn test_memory_feedback_body_serialization() {
2081        use crate::types::{FeedbackSignal, MemoryFeedbackBody};
2082        let body = MemoryFeedbackBody {
2083            agent_id: "agent-1".to_string(),
2084            signal: FeedbackSignal::Flag,
2085        };
2086        let json = serde_json::to_value(body).unwrap();
2087        assert_eq!(json["agent_id"], "agent-1");
2088        assert_eq!(json["signal"], "flag");
2089    }
2090
2091    #[test]
2092    fn test_feedback_memory_url_pattern() {
2093        let client = DakeraClient::new("http://localhost:3000").unwrap();
2094        let memory_id = "mem-abc";
2095        let expected_post = "http://localhost:3000/v1/memories/mem-abc/feedback";
2096        let actual_post = format!("{}/v1/memories/{}/feedback", client.base_url, memory_id);
2097        assert_eq!(actual_post, expected_post);
2098
2099        let expected_patch = "http://localhost:3000/v1/memories/mem-abc/importance";
2100        let actual_patch = format!("{}/v1/memories/{}/importance", client.base_url, memory_id);
2101        assert_eq!(actual_patch, expected_patch);
2102    }
2103
2104    #[test]
2105    fn test_feedback_health_url_pattern() {
2106        let client = DakeraClient::new("http://localhost:3000").unwrap();
2107        let agent_id = "agent-1";
2108        let expected = "http://localhost:3000/v1/feedback/health?agent_id=agent-1";
2109        let actual = format!(
2110            "{}/v1/feedback/health?agent_id={}",
2111            client.base_url, agent_id
2112        );
2113        assert_eq!(actual, expected);
2114    }
2115
2116    // ODE-2 tests
2117    #[test]
2118    fn test_ode_extract_entities_requires_ode_url() {
2119        // Client without ode_url should return Config error.
2120        let client = DakeraClient::new("http://localhost:3000").unwrap();
2121        let rt = tokio::runtime::Runtime::new().unwrap();
2122        let result = rt.block_on(client.ode_extract_entities(ExtractEntitiesRequest {
2123            content: "Alice lives in Paris.".to_string(),
2124            agent_id: "agent-1".to_string(),
2125            memory_id: None,
2126            entity_types: None,
2127        }));
2128        assert!(result.is_err());
2129        let err = result.unwrap_err();
2130        assert!(matches!(err, ClientError::Config(_)));
2131    }
2132
2133    #[test]
2134    fn test_ode_extract_entities_url_built_from_ode_url() {
2135        // Verify the ODE URL is used, not base_url.
2136        let client = DakeraClient::builder("http://localhost:3000")
2137            .ode_url("http://localhost:8080")
2138            .build()
2139            .unwrap();
2140        assert_eq!(client.ode_url.as_deref(), Some("http://localhost:8080"));
2141        let expected = "http://localhost:8080/ode/extract";
2142        let actual = format!("{}/ode/extract", client.ode_url.as_deref().unwrap());
2143        assert_eq!(actual, expected);
2144    }
2145
2146    #[test]
2147    fn test_extract_entities_request_serialization() {
2148        let req = ExtractEntitiesRequest {
2149            content: "Alice in Wonderland".to_string(),
2150            agent_id: "agent-42".to_string(),
2151            memory_id: Some("mem-001".to_string()),
2152            entity_types: Some(vec!["person".to_string(), "location".to_string()]),
2153        };
2154        let json = serde_json::to_string(&req).unwrap();
2155        assert!(json.contains("\"content\":\"Alice in Wonderland\""));
2156        assert!(json.contains("\"agent_id\":\"agent-42\""));
2157        assert!(json.contains("\"memory_id\":\"mem-001\""));
2158        assert!(json.contains("\"person\""));
2159    }
2160
2161    #[test]
2162    fn test_extract_entities_request_omits_none_fields() {
2163        let req = ExtractEntitiesRequest {
2164            content: "hello".to_string(),
2165            agent_id: "a".to_string(),
2166            memory_id: None,
2167            entity_types: None,
2168        };
2169        let json = serde_json::to_string(&req).unwrap();
2170        assert!(!json.contains("memory_id"));
2171        assert!(!json.contains("entity_types"));
2172    }
2173
2174    #[test]
2175    fn test_ode_entity_deserialization() {
2176        let json = r#"{"text":"Alice","label":"person","start":0,"end":5,"score":0.97}"#;
2177        let entity: OdeEntity = serde_json::from_str(json).unwrap();
2178        assert_eq!(entity.text, "Alice");
2179        assert_eq!(entity.label, "person");
2180        assert_eq!(entity.start, 0);
2181        assert_eq!(entity.end, 5);
2182        assert!((entity.score - 0.97).abs() < 1e-4);
2183    }
2184
2185    #[test]
2186    fn test_extract_entities_response_deserialization() {
2187        let json = r#"{
2188            "entities": [
2189                {"text":"Alice","label":"person","start":0,"end":5,"score":0.97},
2190                {"text":"Paris","label":"location","start":16,"end":21,"score":0.92}
2191            ],
2192            "model": "gliner-multi-v2.1",
2193            "processing_time_ms": 34
2194        }"#;
2195        let resp: ExtractEntitiesResponse = serde_json::from_str(json).unwrap();
2196        assert_eq!(resp.entities.len(), 2);
2197        assert_eq!(resp.entities[0].text, "Alice");
2198        assert_eq!(resp.model, "gliner-multi-v2.1");
2199        assert_eq!(resp.processing_time_ms, 34);
2200    }
2201}