atheneum 0.6.2

Agent coordination graph database - episodic and semantic memory for multi-agent workflows
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
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013
1014
1015
1016
1017
1018
1019
1020
1021
1022
1023
1024
1025
1026
1027
# Atheneum Manual

## Installation

### From crates.io

```bash
cargo add atheneum
```

### From source

```bash
git clone https://github.com/oldnordic/atheneum
cd atheneum
cargo build --release
```

---

## Overview

Atheneum is an embedded graph database for AI agent coordination. It stores discoveries, decisions, session histories, task handoffs, and knowledge across agent sessions — replacing ad-hoc file dumps with a queryable, persistent graph.

It is used as a library (embedded in your agent runtime) or accessed via envoy's HTTP bridge (`GET/POST /atheneum/*`).

---

## Opening a Graph

```rust
use atheneum::AtheneumGraph;
use std::path::Path;

// Persistent — creates file if absent
let graph = AtheneumGraph::open(Path::new("atheneum.db"))?;

// In-memory — for tests and ephemeral sessions
let graph = AtheneumGraph::open_in_memory()?;
```

The database schema is auto-migrated on `open()`. No separate migration step required.

---

## Configuration

Atheneum reads `~/.config/atheneum/config.toml` (or `$XDG_CONFIG_HOME/atheneum/config.toml`). A missing file is not an error — sensible defaults are used.

### Default config file

```toml
[atheneum]
db = "~/.local/share/atheneum/atheneum.db"
meta_db = "~/.local/share/atheneum/meta.db"

[llm]
provider = "ollama"
base_url = "http://localhost:11434"
model = "codellama"
api_key = ""

[embeddings]
provider = "hash"
dimension = 128
base_url = "http://localhost:11434"
model = "nomic-embed-text"
api_key = ""

[integrations]
# Cross-tool integration is opt-in. Each tool stays standalone by default.
[integrations.magellan]
enabled = false
config = "~/.config/magellan/config.toml"

[integrations.envoy]
enabled = false
url = "http://localhost:9876"
```

### CLI

```bash
# Create the default config file (idempotent; use --force to overwrite)
atheneum config init

# Print the currently effective configuration as JSON
atheneum config show
```

### Library

```rust
use atheneum::{Config, load_config, save_config};

let cfg = load_config()?;                         // from default location
let path = cfg.db_path();                         // tilde-expanded PathBuf
let meta = cfg.meta_db_path();

save_config(&Config::default())?;                 // write defaults to disk
```

Environment overrides follow the convention `ATHENEUM_<SECTION>_<KEY>` where supported by callers. Paths may contain leading `~`, which is expanded via `$HOME`.

The meta.db routing layer (`MetaRouter::open()`) honors `atheneum.meta_db` from this config and falls back to the XDG default if the config is missing or invalid.

---

## Maintainer Checklist

When changing Atheneum itself, keep the local docs and gates in sync:

```bash
# Repo-local wrappers around the shared project standards
.claude/scripts/quality-gate.sh
printf '{}\n' | env CLAUDE_PROJECT_DIR="$PWD" fish .claude/hooks/verify-rust.fish
bash .claude/hooks/pre-commit-rust-standards
```

Rules:

- Update `CHANGELOG.md` for every user-visible fix, behavior change, or workflow change.
- Update `MANUAL.md` when you add or change a public function, CLI command, flag, or operator workflow.
- Prefer adding the manual/changelog update in the same patch as the code change so the docs cannot drift.
- If a repo-local `.claude/` wrapper exists, run it from the repo root instead of reaching for a shared path manually.

---

## Agent Sessions

Sessions track every coding session — who, when, what branch, how many tool calls, cost.

```rust
use atheneum::graph::{AtheneumGraph, SessionParams};

graph.record_session(SessionParams {
    session_id: "abc-123".into(),
    agent_name: "claude-main".into(),
    project: "my-project".into(),
    tool: "claude-code".into(),
    trigger: "cli".into(),           // "cli" | "subagent" | "hook"
    model: Some("claude-sonnet-4".into()),
    git_branch: Some("feat/auth".into()),
    git_head: Some("a1b2c3d".into()),
    parent_session_id: None,          // set for subagents
})?;
```

### Ending a Session

```rust
use atheneum::graph::EndSessionParams;

graph.end_session(EndSessionParams {
    session_id: "abc-123".into(),
    exit_status: "end_turn".into(),
    prompt_count: 12,
    tool_call_count: 47,
    file_write_count: 3,
    commit_count: 1,
    test_run_count: 2,
    total_input_tokens: 50_000,
    total_output_tokens: 8_000,
    total_cost_usd: 0.15,
})?;
```

### Querying Recent Sessions

```rust
// Last 3 sessions for a project (newest first)
let sessions = graph.query_sessions("my-project", 3, None)?;

// Children of a specific session
let children = graph.query_sessions("my-project", 10, Some("parent-session-id"))?;

for s in sessions {
    println!("{} {} {}tc {}fw last:{:?}",
        s.started_at, s.git_branch.unwrap_or_default(),
        s.tool_call_count, s.file_write_count, s.last_tool);
}
```

### Runtime Cache Stats

Atheneum now keeps a process-local concurrent query cache for the hottest repeated read paths. You can inspect that runtime state directly:

```rust
let stats = graph.runtime_stats();
println!(
    "hits={} misses={} memory_q={} session_q={} wiki_q={}",
    stats.cache_hits,
    stats.cache_misses,
    stats.memory_queries,
    stats.session_queries,
    stats.wiki_queries,
);
```

Current cached reads:

- `query_memory()` / `list_memory()`
- `query_sessions()`
- `query_events()`
- `query_knowledge()` / `query_knowledge_in_project()`
- `list_wiki_pages()`

Writes invalidate the relevant cache domain automatically after successful mutation.

### Tool Call Evidence

```rust
use atheneum::graph::ToolCallParams;

graph.record_evidence_tool_call(ToolCallParams {
    session_id: "abc-123".into(),
    tool_name: "Edit".into(),
    tool_version: None,
    input_hash: Some("deadbeef".into()),
    input_summary: Some("write src/lib.rs".into()),
    output_hash: None,
    output_summary: Some("ok".into()),
    exit_status: "success".into(),
    latency_ms: 234,
    input_tokens_est: None,
    tool_category: "file_write".into(),
})?;
```

### Subagent Handover

```rust
// Subagent writes this on stop — the parent reads it
graph.record_subagent_handover(
    "sub-session-id",
    "Fixed SQL param ordering in query_sessions. evidence.rs line 547.",
    &["src/graph/evidence.rs".to_string()],
    "end_turn",
)?;
```

---

## Discoveries

Discoveries are non-obvious facts, invariants, and decisions stored so future agents don't re-discover them.

```rust
use serde_json::json;

let id = graph.store_discovery(
    "claude",           // agent name
    "Bug",              // discovery type
    "query_sessions",   // target symbol
    json!({
        "file": "src/graph/evidence.rs",
        "line": 547,
        "why": "anonymous ? params required when project is None and parent_id is Some",
        "project_id": "atheneum"
    }),
)?;
```

### Querying Discoveries

```rust
// By target symbol
let discoveries = graph.query_discoveries("query_sessions")?;

// By project (no target required — for session bootstrap context injection)
let recent = graph.recent_project_context("atheneum", 8)?;
```

### Preview Candidate Matches

For fuzzy identifiers, Atheneum can return ranked existing candidates without mutating the graph:

```rust
let candidates = graph.preview_entity_candidates(
    "HTTP Router",
    5,
    Some("atheneum"),
    Some("WikiPage"),
    0.2,
)?;

for candidate in candidates {
    println!("{} {} {:.3}", candidate.kind, candidate.name, candidate.score);
}
```

This is intended for preview/disambiguation flows where you want to inspect likely matches before storing new memory, discovery, or wiki links.

### Query Validation And Repair

Atheneum can preview a navigation query plan before execution:

```rust
let plan = graph.preview_navigate_query(
    "timezone",
    5,
    2,
    None,
    Some("memories"),
)?;

assert!(plan.executable);
assert_eq!(plan.resolved_kind.as_deref(), Some("Memory"));
assert!(plan.kind_repaired);
```

This plan stage:

- trims accidental whitespace from the query
- resolves common entity-kind aliases such as `memory`, `memories`, `wiki`, and `discoveries`
- rejects unknown kinds before traversal instead of silently returning empty results
- records warnings/errors so repaired execution is explicit to callers

### Preview Before Commit

Atheneum can also preview normalized discovery, memory, and handoff payloads before writing:

```rust
let discovery = graph.preview_discovery(
    "codex",
    "pattern",
    "query_cache",
    serde_json::json!({"summary": "cache repeated reads", "project_id": "atheneum"}),
    5,
    0.2,
)?;

let memory = graph.preview_memory(
    "timezone",
    "UTC+1",
    "user",
    0.9,
    None,
    None,
    5,
    0.2,
)?;

let handoff = graph.preview_handoff(
    "claude1",
    "claude2",
    Some("atheneum"),
    serde_json::json!({"task": "finish review", "files_analyzed": ["src/lib.rs"]}),
    5,
    0.2,
)?;
```

These preview APIs:

- do not insert entities or edges
- return deterministic `content_hash` values
- include exact existing matches plus fuzzy candidate matches, even when the fuzzy score alone would have filtered them out

### CLI Navigate Kind Filters

The CLI `navigate` command now accepts `--kind` and reports the repaired/validated plan in its JSON output:

```bash
atheneum navigate ./atheneum.db timezone --kind memories
```
- are intended for operator review or agent-side "propose first, commit later" flows

---

## Knowledge Graph

```rust
// Store a linked discovery
let id = graph.store_discovery_in_project(
    "claude", "Decision", "auth-middleware",
    Some("my-project"),
    json!({ "why": "legal compliance", "risk": "high" }),
)?;

// Query knowledge for a symbol+project
let knowledge = graph.query_knowledge_in_project("auth-middleware", Some("my-project"))?;
```

---

## Task Planning

```rust
use atheneum::graph::AtheneumGraph;
use serde_json::json;

// Create a task
let task_id = graph.create_task("Implement session handover", Some("my-project"))?;

// Add requirements
graph.add_requirement(task_id, "Writes git diff on stop", None)?;

// Update status
graph.update_task_status(task_id, atheneum::graph::KanbanStatus::InProgress)?;
```

---

## Wiki Ingestion

Atheneum parses Markdown files with frontmatter and `[[wikilinks]]` into the knowledge graph.

```rust
let content = r#"---
title: "Session Accountability"
type: concept
---
# Session Accountability
See also [[envoy]] and [[grounded-coding]].
"#;

let entity_id = graph.ingest_wiki_page("session-accountability.md", content, None)?;
```

### Journal Sections

```rust
// Journals use ## HH:MM | Title headers and Kanban lines
let journal = r#"
## 14:23 | Fixed param bug
Corrected SQL ordering in evidence.rs.

## 15:00 | Deployed
"envoy" -> DONE
"#;
let sections = graph.parse_journal_sections(journal)?;
graph.ingest_journal_sections(&sections, Some("my-project"))?;
```

### Searching Wiki Pages

Atheneum uses an FTS5 index over `wiki_pages` for full-text search. Results are ranked by BM25 and include an excerpt only; the full body is never returned by the search API, so you can safely feed results into a context window without accidentally dumping entire articles.

```rust
let hits = graph.search_wiki_pages("session accountability", Some("my-project"), 0, 10)?;
for hit in &hits {
    println!("{} (score={}): {}", hit.path, hit.score, hit.excerpt);
}
```

### Backfilling Wiki Pages

If wiki pages were inserted directly into the `wiki_pages` SQL table (for example, by an older helper script), they may exist as rows but not as proper `WikiPage` graph entities with wikilink edges. `backfill_wiki_pages_to_graph` re-ingests each row through `ingest_wiki_page`, repairing stubs and restoring navigation.

```rust
let fixed = graph.backfill_wiki_pages_to_graph(Some("my-project"))?;
println!("repaired {} pages", fixed.len());
```

### FTS Index Resilience

The `wiki_pages_fts` FTS5 virtual table can be left internally inconsistent when an external SQLite writer (system `sqlite3`, Python, another tool) touches the database between atheneum runs. `AtheneumGraph::open()` detects this on every open and self-heals before creating the connection pool:

1. Probes `wiki_pages_fts` on a fresh connection.
2. If the probe fails, purges the virtual-table entry and shadow tables directly from `sqlite_master` with `PRAGMA writable_schema=ON`.
3. Recreates the table and triggers on another fresh connection.
4. Runs `delete-all` → repopulate from `wiki_pages` → `rebuild` on a fourth fresh connection to finalize shadow-table invariants.
5. Checkpoints WAL so the pool connections open onto a consistent DB.

After healing, `sync-wiki`, `search-wiki`, and `backfill-wiki` work normally. The process is idempotent: a healthy index passes the probe and skips all destructive steps. You do not need to run any manual Python repair script.

---

## HopGraph

HopGraph is atheneum's retrieval model: **embeddings find the door, graph walk retrieves the room.** Unlike flat RAG, HopGraph uses vector similarity only to locate entry points, then expands connected knowledge via graph traversal.

### Token-Budgeted Retrieval

```rust
use atheneum::graph::{AtheneumGraph, EdgeType};

let graph = AtheneumGraph::open(Path::new("atheneum.db"))?;

let views = graph.hopgraph_query(
    "session accountability",      // query text
    3,                             // k: max entry-point entities
    2,                             // depth: BFS expansion depth
    Some(&[EdgeType::Explains, EdgeType::Wikilink]),  // allowed edge types
    2000,                          // max_tokens budget per view
    None,                          // project_id filter
)?;

for view in &views {
    println!("entry={} entities={} edges={}",
        view.entry_id, view.entities.len(), view.edges.len());
}
```

`hopgraph_query` performs: lexical search → filtered BFS subgraph → token-budgeted truncation. Orphan edges (pointing to removed entities) are dropped. The entry entity is always kept regardless of budget.

### Filtered Subgraph Walk

```rust
use atheneum::graph::EdgeType;

// Walk only Explains and Wikilink edges from an entity
let view = graph.get_subgraph_filtered(
    entity_id,
    3,      // depth
    Some(&[EdgeType::Explains, EdgeType::Wikilink]),
)?;
```

### Embedding Backends

```rust
// Default: HashEmbedder (128-dim, zero deps, always available)
let dim = graph.embedder_dimension(); // 128

// Switch to neural embeddings (requires --features neural-embed)
#[cfg(feature = "neural-embed")]
{
    use atheneum::graph::OllamaEmbedder;
    graph.set_embedder(Box::new(OllamaEmbedder::nomic_embed_text()));
    graph.build_search_index()?; // rebuild index with new dimension (768)
    assert_eq!(graph.embedder_dimension(), 768);
}
```

| Backend | Dimension | Dependencies | Quality |
|---------|-----------|-------------|---------|
| `HashEmbedder` | 128 | None | Token overlap only ("car" ≠ "automobile") |
| `OllamaEmbedder` | 768 | ollama + nomic-embed-text | Semantic similarity |

### Discovery Consolidation

Merge duplicate Discovery entities into deduplicated Knowledge entities:

```rust
// Consolidate a single target
let knowledge_id = graph.consolidate_discoveries("query_sessions", Some("my-project"))?;

// Consolidate all targets in a project
let results = graph.consolidation_pass(Some("my-project"))?;
for (target, kid) in &results {
    println!("{} → knowledge {}", target, kid);
}
```

Consolidation creates `DerivedFrom` edges from Knowledge → source Discoveries. Idempotent — re-running returns the existing Knowledge entity.

### Bridge Wiki to Code Symbols

```rust
graph.link_wiki_to_symbols(
    "/path/to/.magellan/magellan/magellan.db",
    "claude",
    Some("my-project"),
)?;
```

For each wiki page's `[[wikilinks]]`, queries the magellan DB for matching code symbols, imports them as Discovery entities, and creates `Explains` edges from wiki page → symbol. Idempotent.

---

## Search

```rust
// Full-text search
let results = graph.full_text_search("query_sessions")?;

// Lexical search via HNSW hash-projected index.
// Matches on shared tokens — not neural/semantic. "car" won't match "automobile".
let results = graph.lexical_search("SQL parameter ordering bug", 5, Some("atheneum"), None, None)?;

// Token-budgeted search — truncate results to fit a context window.
let results = graph.lexical_search("SQL parameter ordering bug", 5, Some("atheneum"), None, Some(500))?;
```

---

## Memory

Memory entries are stable facts stored distinct from Knowledge (merged discoveries) and WikiPage (documents). Each memory has a key, scope, confidence score, and optional project.

Scopes: `user` (preferences), `project` (project facts), `agent` (agent behavior), `memory` (general notes).

Memories are upserted -- storing with the same key, scope, and project_id updates the existing entry instead of creating a duplicate.

```rust
use atheneum::AtheneumGraph;
use std::path::Path;

let graph = AtheneumGraph::open(Path::new("atheneum.db"))?;

// Store a memory
let id = graph.store_memory(
    "timezone",           // key
    "UTC+1",              // content
    "user",               // scope
    0.9,                  // confidence (0.0-1.0)
    None,                 // project_id
    None,                 // tags
)?;

// Retrieve by key
let items = graph.query_memory("timezone", Some("user"), None)?;

// List all memories in a scope
let all = graph.list_memory(Some("user"), None)?;
```

---

## Dream

Dream is atheneum's reflective consolidation pass. It scans memories for problems -- duplicates, stale entries, contradictions, and verbosity -- and either reports them (dry run) or merges them (auto-merge).

What dream does:
1. **SCAN** -- reads all memories in scope
2. **DEDUPLICATE** -- finds near-duplicates using trigram Jaccard similarity (entries that say the same thing differently)
3. **STALE** -- flags entries not updated in N days with low confidence
4. **CONTRADICTION** -- detects same key across different scopes with low content similarity
5. **VERBOSE** -- scores content length vs unique-word ratio
6. **CONSOLIDATED** -- merges findings, creates `SupersededBy` edges pointing old entries to replacements

There are two dream commands:
- `dream` -- runs consolidation over memory entries
- `wiki-dream` -- runs the same pipeline over wiki page entities

```rust
use atheneum::{AtheneumGraph, DreamConfig, DreamMode};
use std::path::Path;

let graph = AtheneumGraph::open(Path::new("atheneum.db"))?;

// Dry run -- report only, no mutations
let report = graph.dream_pass(
    DreamMode::DryRun,
    None,                   // scope filter (None = all)
    Some("my-project"),     // project filter
    &DreamConfig::default(),
)?;
for finding in &report.findings {
    println!("{:?}: {}", finding.phase, finding.description);
}

// Auto-merge -- actually create SupersededBy edges
let report = graph.dream_pass(DreamMode::AutoMerge, None, None, &DreamConfig::default())?;

// Wiki dream -- same pipeline for wiki pages
let wiki_report = graph.wiki_dream_pass(DreamMode::AutoMerge, Some("my-project"), &DreamConfig::default())?;
```

---

## CLI Commands

### Ingest

```bash
# Initialize a new graph database
atheneum init <db-path>

# Sync a wiki directory into the graph
atheneum sync-wiki <db-path> <wiki-dir> [project-id]

# Sync journal files
atheneum sync-journal <db-path> <journal-dir> [project-id]

# Recursively sync a Logseq graph root
atheneum sync-logseq <db-path> <wiki-root> [project-id]

# Import a Claude Code transcript JSONL
atheneum sync-claude-transcript <db-path> <transcript.jsonl> [project-id] [agent-name]

# Store a discovery
atheneum store-discovery <db-path> <agent> <type> <target> [metadata.json]

# Create a relation between two entities
atheneum add-edge <db-path> <from-id> <to-id> <edge-type> [data.json|--data 'json']
```

`sync-logseq` expects a Logseq-style root with `pages/` and/or `journals/`. It recursively ingests markdown files under those directories. Wiki page `[[links]]` are stored as first-class `wikilink` edges, enabling graph traversal through article and note relationships.

`sync-claude-transcript` expects a Claude Code transcript JSONL, typically under `~/.claude/projects/<encoded-project>/<session-id>.jsonl`. It imports prompt summaries, assistant replies, observed tool calls, `accessed` file relations for `Read`/`Edit`/`Write`, and session token/cache totals. Re-running on the same append-only transcript imports only new lines because Atheneum stores a transcript cursor in SQL.

`store-discovery` takes an optional JSON file for metadata. The metadata JSON can contain fields like `project_id`, `why`, `file`, `line`.

`add-edge` creates a typed edge between two entities. Valid edge types include: `performed_by`, `assigned_to`, `called`, `accessed`, `modified`, `verified_by`, `caused_by`, `created`, `related_to`, `mentions`, `wikilink`, `implements`, `depends_on`, `tested_by`, `fixed_by`, `regressed_by`, `observed_in`, `belongs_to_project`, `similar_failure`, `requires_skill`, `handled_by_tool`, `explains`, `derived_from`, `superseded_by`, `consolidated_from`.

### Tasks

```bash
# Create a new task
atheneum task-create <db-path> <title> [description] [--project P]

# List tasks (default: non-archived)
atheneum task-list <db-path> [--project P] [--status S]

# List archived tasks explicitly
atheneum task-list <db-path> --status ARCHIVED [--project P]

# Update task status
atheneum task-update <db-path> <task-id> <status>

# Mark task as DONE
atheneum task-done <db-path> <task-id>

# Archive a task
atheneum task-archive <db-path> <task-id>
```

Valid statuses: `TODO`, `IN_PROGRESS`, `DONE`, `BLOCKED`, `ARCHIVED`.

### Memory

```bash
# Store a memory
atheneum memory-store <db-path> <key> <content> [--scope S] [--confidence N] [--project P]

# Retrieve memory by key
atheneum memory-get <db-path> <key> [--scope S] [--project P]

# List memories (paginated; default limit 1000)
atheneum memory-list <db-path> [--scope S] [--project P] [--offset N] [--limit N]
```

Memories are upserted -- storing with the same key + scope + project updates the existing entry. Default scope is `user`, default confidence is `1.0`.

### Dream

```bash
# Run reflective memory consolidation pass
atheneum dream <db-path> [--scope S] [--project P] [--dry-run|--auto-merge]

# Run consolidation over wiki pages
atheneum wiki-dream <db-path> [--project P] [--dry-run|--auto-merge]
```

`--dry-run` (default) reports findings without modifying the graph. `--auto-merge` creates `SupersededBy` edges pointing old entries to their replacements.

Output is a JSON `DreamReport` with findings organized by phase (DEDUPLICATE, STALE, CONTRADICTION, VERBOSE, CONSOLIDATED).

### Query and Navigation

```bash
# HNSW/lexical search over all entities
atheneum search <db-path> <query> [--k N] [--project P] [--max-tokens N]

# Search then BFS-walk graph subgraphs
atheneum navigate <db-path> <query> [--k N] [--depth N] [--project P] [--kind K] [--max-tokens N] [--concise]

# Query a wiki page by path
atheneum query-wiki <db-path> <path>

# Full-text search over wiki pages (excerpts only; no full body)
atheneum search-wiki <db-path> <query> [--limit N] [--offset N] [--project P]

# Backfill wiki pages written directly to the SQL table into the graph
atheneum backfill-wiki <db-path> [--project P]

# Query journal sections by path
atheneum query-journal <db-path> <path>

# Aggregated knowledge for a target
atheneum query-knowledge <db-path> <target> [--project P] [--max-tokens N]

# Session history
atheneum query-sessions <db-path> [--project P] [--offset N] [--limit N]

# Event log
atheneum query-events <db-path> [--session <id>] [--type <type>] [--offset N] [--limit N]

# List wiki pages (default limit 1000)
atheneum list-pages <db-path> [--project P] [--offset N] [--limit N]

# Print a graph entity as JSON
atheneum entity <db-path> <entity-id>

# Print a graph edge as JSON
atheneum edge <db-path> <edge-id>

# One-hop edges or BFS subgraph
atheneum neighbors <db-path> <entity-id> [--depth N]

# Graph topology counts
atheneum graph-stats <db-path>
```

`search` uses the HNSW lexical index. It matches on shared tokens -- not semantic similarity. "car" will not match "automobile". Good for symbol and identifier search. Use `--max-tokens` to truncate the result list before it reaches your LLM context window.

`search-wiki` uses the FTS5 index over `wiki_pages` (`title`, `body`, and `path`). It returns ranked excerpts only; the full article body is never included in the output. Prefix queries work automatically: searching `rout` matches `Router`, `Routes`, and path fragments like `wiki/router.md`. If FTS5 returns no hits, `search-wiki` falls back to a graph-entity name/path/title substring search so partial concept queries still find stored pages. Use `--limit` and `--offset` for pagination, and `--project` to scope the search.

`list-pages` returns metadata for every wiki page (path, title, project, timestamps) without requiring a query. Use it to browse what is stored before searching, or to enumerate pages for export.

`backfill-wiki` re-ingests every `wiki_pages` SQL row through `ingest_wiki_page`. Use it to repair pages that were written directly to the SQL table without creating a proper `WikiPage` graph entity or wikilink edges. It skips pages whose graph entity already has a body and is not marked as a stub.

`navigate` performs a search, then expands each hit into a subgraph using BFS. The `--kind` flag filters by entity type (accepts aliases like `memory`, `memories`, `wiki`, `discoveries`). The output includes the validated query plan plus subgraph views. Use `--max-tokens` to truncate each subgraph view to a token budget (the entry entity is always kept; neighbors are dropped until the budget fits). Use `--concise` to emit compact Markdown instead of JSON — designed for pasting into a language-model context window.

`query-knowledge` aggregates discoveries and handoffs for a target. Use `--max-tokens` to limit the total response size; discoveries are dropped first, then handoffs, and `"truncated": true` is set when truncation occurs.

### Cross-Project Registry (Meta)

```bash
# Register a project in the meta.db routing layer
atheneum meta-register envoy /home/feanor/Projects/envoy \
  /home/feanor/Projects/envoy/.magellan/magellan.db \
  --atheneum-db /home/feanor/Projects/envoy/atheneum.db \
  --language rust

# List all registered projects
atheneum meta-list

# List projects filtered by language
atheneum meta-list --language rust
```

`meta-register` upserts a project into `~/.local/share/atheneum/meta.db` (or `$XDG_DATA_HOME/atheneum/meta.db`). Re-registering the same name updates all fields and re-enables the project.

`meta-list` queries enabled projects from the registry. Use `--language` to filter by programming language.

### Cross-Project Queries

Atheneum can query across magellan-indexed codebases without importing their data. It uses `meta.db` as a routing registry and lazily `ATTACH DATABASE` each project's magellan DB on demand.

```bash
# Search for a symbol across all Rust projects
atheneum cross-search "build_router" --language rust --k 10

# Search across all registered projects (no language filter)
atheneum cross-search "checkpoint" --k 20

# Navigate: search + BFS subgraph walk per project
atheneum cross-navigate "error handling" --language rust --k 5 --depth 2
```

Output is JSON. `cross-search` returns ranked symbol hits with project, name, kind, and file path. `cross-navigate` returns one subgraph view per entry point, including entities and edges from each attached magellan database.

The router keeps an LRU cache of attached databases (default capacity 8). Missing or unreadable databases are skipped with a warning rather than aborting the whole query.

#### End-to-End Example: Finding All HTTP Router Implementations

Imagine you maintain three Rust projects — `envoy`, `magellan`, and `atheneum` — and you want to see how each one implements its HTTP router. Here is the complete workflow:

**Step 1 — Index each project with magellan (one-time per project):**

```bash
cd ~/Projects/envoy
magellan watch --root ./src --db ~/.magellan/envoy/envoy.db --scan-initial

cd ~/Projects/magellan
magellan watch --root ./src --db ~/.magellan/magellan/magellan.db --scan-initial

cd ~/Projects/atheneum
magellan watch --root ./src --db ~/.magellan/atheneum/atheneum.db --scan-initial
```

**Step 2 — Register each project in atheneum's meta.db (one-time per project):**

```bash
atheneum meta-register envoy ~/Projects/envoy \
  ~/.magellan/envoy/envoy.db --language rust

atheneum meta-register magellan ~/Projects/magellan \
  ~/.magellan/magellan/magellan.db --language rust

atheneum meta-register atheneum ~/Projects/atheneum \
  ~/.magellan/atheneum/atheneum.db --language rust
```

**Step 3 — Search across all three projects for "build_router":**

```bash
atheneum cross-search "build_router" --language rust --k 10
```

Sample output:

```json
{
  "results": [
    {
      "project": "envoy",
      "name": "build_router",
      "kind": "Function",
      "file": "src/server.rs",
      "line": 42,
      "score": 1.0
    },
    {
      "project": "magellan",
      "name": "build_router",
      "kind": "Function",
      "file": "src/http/mod.rs",
      "line": 88,
      "score": 1.0
    }
  ]
}
```

**Step 4 — Navigate deeper: see what each router calls and what calls it:**

```bash
atheneum cross-navigate "build_router" --language rust --k 3 --depth 2
```

This returns one subgraph per entry-point match. Each subgraph shows the function's callers, callees, and related symbols — directly from each project's magellan DB, without copying data into atheneum.

**Step 5 — Use in a script or agent:**

```rust
use atheneum::CrossRouter;

fn main() -> anyhow::Result<()> {
    let mut router = CrossRouter::open()?;

    // Find all Rust projects that have a "build_router" function
    let hits = router.cross_search("build_router", Some("rust"), 10)?;
    for hit in &hits {
        println!("{}: {} in {}:{}",
            hit.project, hit.name, hit.file, hit.line.unwrap_or(0));
    }

    // For each hit, expand 2 hops of graph context
    let views = router.cross_navigate("build_router", Some("rust"), 3, 2)?;
    for view in &views {
        println!("project={} entities={} edges={}",
            view.project, view.subgraph.entities.len(), view.subgraph.edges.len());
    }

    Ok(())
}
```

**What happens under the hood:**

1. `CrossRouter::open()` reads `~/.config/atheneum/config.toml` to find `meta_db` (or uses the XDG default).
2. `cross_search` queries the `project_registry` table for enabled Rust projects, then `ATTACH DATABASE`es each magellan DB one at a time (or reuses cached attachments).
3. Each attached DB is queried via a cross-schema `UNION ALL` over `graph_entities` and `graph_edges`.
4. Results are ranked by exact-match score and returned as a single list.
5. `cross_navigate` does the same search, then runs BFS per entry point per project, returning subgraph views with full entity/edge detail.

**Cleaning up:**

```bash
# Remove a project from the registry (soft-disable)
atheneum meta-register envoy ... --disable

# Re-enable
atheneum meta-register envoy ...  # omit --disable
```

### Config

```bash
# Create the default config file at ~/.config/atheneum/config.toml
atheneum config init

# Overwrite an existing config file
atheneum config init --force

# Print the effective configuration as JSON
atheneum config show
```

`config init` writes the default TOML (XDG paths, local Ollama defaults, disabled cross-tool integrations). `config show` reads the file (or defaults if missing) and prints JSON, which is useful for debugging path expansion and integration flags.

### Maintenance

```bash
# Rebuild HNSW search index
atheneum reindex <db-path>

# Merge discoveries into Knowledge entities
atheneum consolidate <db-path> [target] [--project P]

# Print version
atheneum --version

# Print help
atheneum help
```

`reindex` rebuilds the HNSW index over all entities and then runs a WAL checkpoint to reclaim disk space. Useful after bulk imports or if search results seem incomplete. (Prior to v0.6.2, the checkpoint call could panic with "Execute returned results"; it now uses `query_row` because `PRAGMA wal_checkpoint` returns a row.)

`consolidate` merges all Discovery entities for a target (or all targets) into deduplicated Knowledge entities with `DerivedFrom` edges. Idempotent -- re-running returns the existing Knowledge entity.

---

## Features

| Feature | Default | Description |
|---------|---------|-------------|
| `default` | yes | Core graph, wiki, sessions, planning, search |
| `neural-embed` | no | Ollama neural embeddings (requires `ureq`, ollama + nomic-embed-text) |
| `web` | no | Web dashboard (axum + askama templates) |
| `cli` | no | `atheneum` CLI binary |
| `async` | no | Async runtime support |

---

## Error Handling

All functions return `anyhow::Result<T>`. Errors include context about which operation failed.

```rust
match graph.record_session(params) {
    Ok(()) => {},
    Err(e) => eprintln!("Session record failed: {:#}", e),
}
```

---

## Thread Safety

`AtheneumGraph` uses internal `Mutex` locking. The `pub` methods take `&self` (shared reference) and handle synchronization internally. For concurrent access from multiple threads, wrap in `Arc<AtheneumGraph>` or use connection pooling per thread.

---

## Requirements

- Rust 1.75+
- SQLite 3.35+ with JSON1 extension (bundled via rusqlite by default)

## License

GPL-3.0-only -- see [LICENSE](LICENSE).