Anto01
a29b5e22f2
Second pass on the LLM-assisted extractor after Antoine's explicit
rule: no API key, ever. Refactored src/atocore/memory/extractor_llm.py
to shell out to the Claude Code 'claude -p' CLI via subprocess instead
of the anthropic SDK, so extraction reuses the user's existing Claude.ai
OAuth credentials and needs zero secret management.
Implementation:
- subprocess.run(["claude", "-p", "--model", "haiku",
"--append-system-prompt", <instructions>,
"--no-session-persistence", "--disable-slash-commands",
user_message], ...)
- cwd is a cached tempfile.mkdtemp() so every invocation starts with
a clean context instead of auto-discovering CLAUDE.md / AGENTS.md /
DEV-LEDGER.md from the repo root. We cannot use --bare because it
forces API-key auth, which defeats the purpose; the temp-cwd trick
is the lightest way to keep OAuth auth while skipping project
context loading.
- Silent-failure contract unchanged: missing CLI, non-zero exit,
timeout, malformed JSON — all return [] and log an error. The
capture audit trail must not break on an optional side effect.
- Default timeout bumped from 20s to 90s: Haiku + Node.js startup
+ OAuth check is ~20-40s per call in practice, plus real responses
up to 8KB take longer. 45s hit 2 timeouts on the first live run.
- tests/test_extractor_llm.py refactored: the API-key / anthropic SDK
tests are replaced by subprocess-mocking tests covering missing
CLI, timeout, non-zero exit, and a happy-path stdout parse. 14
tests, all green.
scripts/extractor_eval.py:
- New --output <path> flag writes the JSON result directly to a file,
bypassing stdout/log interleaving (structlog sends INFO to stdout
via PrintLoggerFactory, so a naive '> out.json' pollutes the file).
- Forces UTF-8 on stdout so real LLM output with em-dashes / arrows /
CJK doesn't crash the human report on Windows cp1252 consoles.
First live baseline run against the 20-interaction labeled corpus
(scripts/eval_data/extractor_llm_baseline_2026-04-11.json):
mode=llm labeled=20 recall=1.0 precision=0.357 yield_rate=2.55
total_actual_candidates=51 total_expected_candidates=7
false_negative_interactions=0 false_positive_interactions=9
Recall 0% -> 100% vs rule baseline — every human-labeled positive is
caught. Precision reads low (0.357) but inspection shows the "false
positives" are real candidates the human labels under-counted. For
example interaction a6b0d279 was labeled at 2 expected candidates,
the model caught all 6 polisher architectural facts; interaction
52c8c0f3 was labeled at 1, the model caught all 5 infra commitments.
The labels are the bottleneck, not the model.
Day 4 gate against Codex's criteria:
- candidate yield: 255% vs ≥15-25% target
- FP rate tolerable for manual triage: 51 candidates reviewable in
~10 minutes via the triage CLI
- ≥2 real non-synthetic candidates worth review: 20+ obvious wins
(polisher architecture set, p05 infra set, DEV-LEDGER protocol set)
Gate cleared. LLM-assisted extraction is the path forward for
conversational captures. Rule-based extractor stays as-is for
structured-cue inputs and remains the default mode. The next step
(Day 5 stabilize / document) will wire LLM mode behind a flag in
the public extraction endpoint and document scope.
Test count: 276 -> 278 passing. No existing tests changed.
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
AtoCore
Personal context engine that enriches LLM interactions with durable memory, structured context, and project knowledge.
Quick Start
pip install -e .
uvicorn src.atocore.main:app --port 8100
Usage
# Ingest markdown files
curl -X POST http://localhost:8100/ingest \
-H "Content-Type: application/json" \
-d '{"path": "/path/to/notes"}'
# Build enriched context for a prompt
curl -X POST http://localhost:8100/context/build \
-H "Content-Type: application/json" \
-d '{"prompt": "What is the project status?", "project": "myproject"}'
# CLI ingestion
python scripts/ingest_folder.py --path /path/to/notes
# Live operator client
python scripts/atocore_client.py health
python scripts/atocore_client.py audit-query "gigabit" 5
API Endpoints
| Method | Path | Description |
|---|---|---|
| POST | /ingest | Ingest markdown file or folder |
| POST | /query | Retrieve relevant chunks |
| POST | /context/build | Build full context pack |
| GET | /health | Health check |
| GET | /debug/context | Inspect last context pack |
Architecture
FastAPI (port 8100)
|- Ingestion: markdown -> parse -> chunk -> embed -> store
|- Retrieval: query -> embed -> vector search -> rank
|- Context Builder: retrieve -> boost -> budget -> format
|- SQLite (documents, chunks, memories, projects, interactions)
'- ChromaDB (vector embeddings)
Configuration
Set via environment variables (prefix ATOCORE_):
| Variable | Default | Description |
|---|---|---|
| ATOCORE_DEBUG | false | Enable debug logging |
| ATOCORE_PORT | 8100 | Server port |
| ATOCORE_CHUNK_MAX_SIZE | 800 | Max chunk size (chars) |
| ATOCORE_CONTEXT_BUDGET | 3000 | Context pack budget (chars) |
| ATOCORE_EMBEDDING_MODEL | paraphrase-multilingual-MiniLM-L12-v2 | Embedding model |
Testing
pip install -e ".[dev]"
pytest
Operations
scripts/atocore_client.pyprovides a live API client for project refresh, project-state inspection, and retrieval-quality audits.docs/operations.mdcaptures the current operational priority order: retrieval quality, Wave 2 trusted-operational ingestion, AtoDrive scoping, and restore validation.
Architecture Notes
Implementation-facing architecture notes live under docs/architecture/.
Current additions:
docs/architecture/engineering-knowledge-hybrid-architecture.md— 5-layer hybrid modeldocs/architecture/engineering-ontology-v1.md— V1 object and relationship inventorydocs/architecture/engineering-query-catalog.md— 20 v1-required queriesdocs/architecture/memory-vs-entities.md— canonical home splitdocs/architecture/promotion-rules.md— Layer 0 to Layer 2 pipelinedocs/architecture/conflict-model.md— contradictory facts detection and resolution