converge-provider 3.2.0

LLM provider implementations for Converge
Documentation
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
// Copyright 2024-2026 Reflective Labs
// SPDX-License-Identifier: MIT

//! Provider contract types for structured observations and call context.
//!
//! These types define the boundary between providers (adapters) and the
//! Converge core engine. Providers produce observations; the engine decides.

use serde::{Deserialize, Serialize};
use std::collections::hash_map::DefaultHasher;
use std::hash::{Hash, Hasher};
use std::time::Instant;

/// Capabilities that providers can offer.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, Serialize, Deserialize)]
pub enum Capability {
    /// LLM text completion
    LlmCompletion,
    /// Text/image embedding generation
    Embedding,
    /// Re-ranking search results
    Reranking,
    /// Vector similarity search
    VectorSearch,
    /// Web search with citations
    WebSearch,
    /// Graph pattern matching
    GraphSearch,
}

/// Data sovereignty region for a provider.
#[derive(Debug, Clone, Copy, PartialEq, Eq, Hash, Serialize, Deserialize)]
pub enum Region {
    /// United States
    US,
    /// European Union
    EU,
    /// China
    CN,
    /// Local (on-premise, no network)
    Local,
}

impl std::fmt::Display for Region {
    fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
        match self {
            Self::US => write!(f, "US"),
            Self::EU => write!(f, "EU"),
            Self::CN => write!(f, "CN"),
            Self::Local => write!(f, "Local"),
        }
    }
}

/// Metadata about a provider implementation.
///
/// This is static information that describes what a provider offers.
#[derive(Debug, Clone)]
pub struct ProviderMeta {
    /// Provider name (e.g., "anthropic", "openai")
    pub name: &'static str,
    /// Provider version
    pub version: &'static str,
    /// Capabilities this provider offers
    pub capabilities: &'static [Capability],
    /// Vendor identifier
    pub vendor: &'static str,
    /// Region (for data sovereignty)
    pub region: Region,
}

impl ProviderMeta {
    /// Create new provider metadata.
    #[must_use]
    pub const fn new(
        name: &'static str,
        version: &'static str,
        capabilities: &'static [Capability],
        vendor: &'static str,
        region: Region,
    ) -> Self {
        Self {
            name,
            version,
            capabilities,
            vendor,
            region,
        }
    }
}

/// Context passed to every provider call for tracing and budgets.
///
/// This provides correlation IDs, timeouts, and budget constraints.
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ProviderCallContext {
    /// Root intent ID for correlation (from converge-core)
    pub root_intent_id: Option<String>,
    /// Trace ID for distributed tracing
    pub trace_id: String,
    /// User/org identifier (for auditing)
    pub user_id: Option<String>,
    /// Maximum allowed latency in milliseconds (timeout)
    pub timeout_ms: u64,
    /// Maximum allowed cost in USD
    pub max_cost: Option<f64>,
    /// Maximum tokens (input + output)
    pub max_tokens: Option<u32>,
}

impl Default for ProviderCallContext {
    fn default() -> Self {
        Self {
            root_intent_id: None,
            trace_id: generate_trace_id(),
            user_id: None,
            timeout_ms: 30_000, // 30 seconds
            max_cost: None,
            max_tokens: None,
        }
    }
}

impl ProviderCallContext {
    /// Create a new call context with a specific trace ID.
    pub fn with_trace_id(trace_id: impl Into<String>) -> Self {
        Self {
            trace_id: trace_id.into(),
            ..Default::default()
        }
    }

    /// Set the root intent ID.
    #[must_use]
    pub fn with_root_intent(mut self, root_intent_id: impl Into<String>) -> Self {
        self.root_intent_id = Some(root_intent_id.into());
        self
    }

    /// Set the user ID.
    #[must_use]
    pub fn with_user(mut self, user_id: impl Into<String>) -> Self {
        self.user_id = Some(user_id.into());
        self
    }

    /// Set the timeout in milliseconds.
    #[must_use]
    pub fn with_timeout_ms(mut self, timeout_ms: u64) -> Self {
        self.timeout_ms = timeout_ms;
        self
    }

    /// Set the maximum cost budget.
    #[must_use]
    pub fn with_max_cost(mut self, max_cost: f64) -> Self {
        self.max_cost = Some(max_cost);
        self
    }

    /// Set the maximum token budget.
    #[must_use]
    pub fn with_max_tokens(mut self, max_tokens: u32) -> Self {
        self.max_tokens = Some(max_tokens);
        self
    }
}

/// Token usage information from a provider call.
#[derive(Debug, Clone, Copy, Serialize, Deserialize)]
pub struct TokenUsage {
    /// Number of input tokens
    pub input_tokens: u32,
    /// Number of output tokens
    pub output_tokens: u32,
}

impl TokenUsage {
    /// Create new token usage.
    #[must_use]
    pub const fn new(input_tokens: u32, output_tokens: u32) -> Self {
        Self {
            input_tokens,
            output_tokens,
        }
    }

    /// Total tokens used.
    #[must_use]
    pub const fn total(&self) -> u32 {
        self.input_tokens + self.output_tokens
    }
}

/// Structured result from every provider call.
///
/// This is the core type that providers return. It includes:
/// - The actual content
/// - Provenance metadata for tracing
/// - Cost and latency information
///
/// # Type Parameter
///
/// `T` is the content type (typically `String` for LLM responses).
#[derive(Debug, Clone, Serialize, Deserialize)]
pub struct ProviderObservation<T> {
    /// Stable reference ID for this observation
    pub observation_id: String,
    /// Canonical hash of the request
    pub request_hash: String,
    /// Provider that produced this observation
    pub vendor: String,
    /// Model used
    pub model: String,
    /// Call latency in milliseconds
    pub latency_ms: u64,
    /// Estimated cost in USD (if known)
    pub cost_estimate: Option<f64>,
    /// Token usage (if applicable)
    pub tokens: Option<TokenUsage>,
    /// The actual content
    pub content: T,
    /// Raw response (optional, size-bounded)
    #[serde(skip_serializing_if = "Option::is_none")]
    pub raw_response: Option<String>,
}

impl<T> ProviderObservation<T> {
    /// Create a new observation.
    pub fn new(
        vendor: impl Into<String>,
        model: impl Into<String>,
        content: T,
        latency_ms: u64,
    ) -> Self {
        let observation_id = generate_observation_id();
        Self {
            observation_id,
            request_hash: String::new(),
            vendor: vendor.into(),
            model: model.into(),
            latency_ms,
            cost_estimate: None,
            tokens: None,
            content,
            raw_response: None,
        }
    }

    /// Set the request hash.
    #[must_use]
    pub fn with_request_hash(mut self, hash: impl Into<String>) -> Self {
        self.request_hash = hash.into();
        self
    }

    /// Set the cost estimate.
    #[must_use]
    pub fn with_cost(mut self, cost: f64) -> Self {
        self.cost_estimate = Some(cost);
        self
    }

    /// Set the token usage.
    #[must_use]
    pub fn with_tokens(mut self, input: u32, output: u32) -> Self {
        self.tokens = Some(TokenUsage::new(input, output));
        self
    }

    /// Set the raw response (will be truncated if too long).
    #[must_use]
    pub fn with_raw_response(mut self, raw: impl Into<String>) -> Self {
        let raw = raw.into();
        const MAX_RAW_SIZE: usize = 10_000;
        if raw.len() > MAX_RAW_SIZE {
            self.raw_response = Some(format!("{}...[truncated]", &raw[..MAX_RAW_SIZE]));
        } else {
            self.raw_response = Some(raw);
        }
        self
    }

    /// Generate provenance string for Facts.
    ///
    /// This string can be attached to `ProposedFact` instances to trace
    /// where the data came from.
    pub fn provenance(&self) -> String {
        format!("{}:{}:{}", self.vendor, self.model, self.observation_id)
    }
}

/// A timer for measuring provider call latency.
///
/// Use this to accurately measure call duration.
pub struct CallTimer {
    start: Instant,
}

impl CallTimer {
    /// Start a new timer.
    #[must_use]
    pub fn start() -> Self {
        Self {
            start: Instant::now(),
        }
    }

    /// Get elapsed time in milliseconds.
    #[must_use]
    pub fn elapsed_ms(&self) -> u64 {
        self.start.elapsed().as_millis() as u64
    }
}

/// Compute a canonical hash for a request.
///
/// This creates a deterministic fingerprint of a request that can be used
/// for caching and provenance tracking.
#[must_use]
pub fn canonical_hash(data: &str) -> String {
    let mut hasher = DefaultHasher::new();
    data.hash(&mut hasher);
    format!("hash:{:016x}", hasher.finish())
}

/// Generate a unique observation ID.
fn generate_observation_id() -> String {
    use std::sync::atomic::{AtomicU64, Ordering};
    static COUNTER: AtomicU64 = AtomicU64::new(0);

    let count = COUNTER.fetch_add(1, Ordering::Relaxed);
    let timestamp = std::time::SystemTime::now()
        .duration_since(std::time::UNIX_EPOCH)
        .map(|d| d.as_millis())
        .unwrap_or(0);

    format!("obs-{timestamp:x}-{count:x}")
}

/// Generate a trace ID.
fn generate_trace_id() -> String {
    use std::sync::atomic::{AtomicU64, Ordering};
    static COUNTER: AtomicU64 = AtomicU64::new(0);

    let count = COUNTER.fetch_add(1, Ordering::Relaxed);
    let timestamp = std::time::SystemTime::now()
        .duration_since(std::time::UNIX_EPOCH)
        .map(|d| d.as_millis())
        .unwrap_or(0);

    format!("trace-{timestamp:x}-{count:x}")
}

#[cfg(test)]
mod tests {
    use super::*;

    #[test]
    fn test_provider_meta() {
        static CAPS: &[Capability] = &[Capability::LlmCompletion];
        let meta = ProviderMeta::new("test", "1.0", CAPS, "test-vendor", Region::US);
        assert_eq!(meta.name, "test");
        assert_eq!(meta.region, Region::US);
    }

    #[test]
    fn test_call_context_default() {
        let ctx = ProviderCallContext::default();
        assert_eq!(ctx.timeout_ms, 30_000);
        assert!(ctx.trace_id.starts_with("trace-"));
    }

    #[test]
    fn test_call_context_builder() {
        let ctx = ProviderCallContext::default()
            .with_root_intent("intent-123")
            .with_user("user-456")
            .with_timeout_ms(5000)
            .with_max_cost(1.0)
            .with_max_tokens(1000);

        assert_eq!(ctx.root_intent_id, Some("intent-123".into()));
        assert_eq!(ctx.user_id, Some("user-456".into()));
        assert_eq!(ctx.timeout_ms, 5000);
        assert_eq!(ctx.max_cost, Some(1.0));
        assert_eq!(ctx.max_tokens, Some(1000));
    }

    #[test]
    fn test_token_usage() {
        let usage = TokenUsage::new(100, 50);
        assert_eq!(usage.total(), 150);
    }

    #[test]
    fn test_observation_provenance() {
        let obs = ProviderObservation::new("anthropic", "claude-3", "content", 100);
        let prov = obs.provenance();
        assert!(prov.starts_with("anthropic:claude-3:obs-"));
    }

    #[test]
    fn test_observation_builder() {
        let obs = ProviderObservation::new("openai", "gpt-4", "response", 500)
            .with_request_hash("hash:abc123")
            .with_cost(0.05)
            .with_tokens(100, 50);

        assert_eq!(obs.request_hash, "hash:abc123");
        assert_eq!(obs.cost_estimate, Some(0.05));
        assert_eq!(obs.tokens.unwrap().total(), 150);
    }

    #[test]
    fn test_raw_response_truncation() {
        let long_response = "x".repeat(20_000);
        let obs = ProviderObservation::new("test", "model", "content", 100)
            .with_raw_response(long_response);

        let raw = obs.raw_response.unwrap();
        assert!(raw.ends_with("...[truncated]"));
        assert!(raw.len() < 15_000);
    }

    #[test]
    fn test_canonical_hash_deterministic() {
        let hash1 = canonical_hash("test input");
        let hash2 = canonical_hash("test input");
        assert_eq!(hash1, hash2);

        let hash3 = canonical_hash("different input");
        assert_ne!(hash1, hash3);
    }

    #[test]
    fn test_call_timer() {
        let timer = CallTimer::start();
        std::thread::sleep(std::time::Duration::from_millis(10));
        let elapsed = timer.elapsed_ms();
        assert!(elapsed >= 10);
    }

    #[test]
    fn test_observation_ids_unique() {
        let obs1 = ProviderObservation::new("test", "model", "a", 1);
        let obs2 = ProviderObservation::new("test", "model", "b", 2);
        assert_ne!(obs1.observation_id, obs2.observation_id);
    }
}