Chronicle Memory

Chronicle Memory is an evidence-only textual memory system for the Agent Memory Challenge 2026. It is designed for the Academic Methods board and exposes the required synchronous Add and Search endpoints.

Method

The system stores each source chunk in SQLite under an explicit user_id and session_id boundary. Search ranks only stored evidence with a transparent hybrid scorer:

1. BM25-style lexical relevance with document-frequency weighting.
2. Exact phrase and token-overlap bonuses for fact and multi-hop questions.
3. Year matching and recency weighting for temporal questions.
4. Optional same-session boost and source-session diversity.
5. Optional gpt-4o-mini annotations/query expansion, used only to improve retrieval terms. The model never writes the final answer and Search never fabricates evidence.

This is an original implementation for this submission. It does not copy a third-party repository or benchmark answer. The platform remains responsible for Answer, Eval, and leaderboard publication.

Run locally

Requires Python 3.11+.

python app.py

The service listens on http://127.0.0.1:8000 by default. Data is stored in data/memories.sqlite3.

To enable the official model path, provide a key outside the repository:

$env:OPENAI_API_KEY = "..."
$env:OPENAI_MODEL = "gpt-4o-mini"
python app.py

Without a key, the deterministic retrieval path remains available for local contract tests. For a formal platform reproduction, configure OPENAI_API_KEY and keep OPENAI_MODEL=gpt-4o-mini.

Docker

docker build -t chronicle-memory:0.1.0 .
docker run --rm -p 8000:8000 -e OPENAI_API_KEY=YOUR_KEY chronicle-memory:0.1.0

Do not commit API keys. The service accepts X-Api-Key, Authorization: Bearer ..., or Authorization: Token ... when MEMORY_API_KEY is configured. If no key is configured, the service is unauthenticated for local smoke testing.

API contract

Health is an unauthenticated GET /health.

Add is synchronous and returns only after persistence:

curl -X POST http://127.0.0.1:8000/add \
  -H 'Content-Type: application/json' \
  -d '{"request_id":"demo-1","user_id":"demo-user","session_id":"demo-session","content":"The launch is on 2026-08-07 in Shanghai."}'

Example response:

{"success":true,"request_id":"demo-1","user_id":"demo-user","session_id":"demo-session","memory_ids":["mem_..."]}

Search returns relevance-ordered memory evidence and never a generated answer:

curl -X POST http://127.0.0.1:8000/search \
  -H 'Content-Type: application/json' \
  -d '{"query":"When and where is the launch?","user_id":"demo-user","top_k":100}'

Example response:

{"data":[{"id":"mem_...","content":"The launch is on 2026-08-07 in Shanghai."}]}

The formal contract uses top_k <= 100. session_id on Search is accepted as an optional local ranking hint, but user_id is the only isolation boundary.

Verification

python -m unittest discover -s tests -v

Manual smoke sequence:

Invoke-RestMethod http://127.0.0.1:8000/health
Invoke-RestMethod -Method Post -Uri http://127.0.0.1:8000/add -ContentType 'application/json' -Body (@{request_id='smoke-1';user_id='smoke-user';session_id='smoke-session';content='Ada reviewed the retrieval design in Oxford in 2026.'} | ConvertTo-Json)
Invoke-RestMethod -Method Post -Uri http://127.0.0.1:8000/search -ContentType 'application/json' -Body (@{query='Where did Ada review the design?';user_id='smoke-user';top_k=100} | ConvertTo-Json)

Submission checklist

• Track: Textual Memory.
• Division: Academic Methods.
• Route: public GitHub repository, platform-deployed Docker code.
• Version: 0.1.0.
• Entrypoint: python app.py or the Docker command above.
• Endpoints: GET /health, POST /add, POST /search.
• Model declaration: gpt-4o-mini is the optional production annotation/query-expansion model; the scoring and evidence boundary are implemented locally.
• Credentials: supply through the evaluation form or deployment environment only.
• Fixed-version rule: tag the repository before requesting formal Full evaluation and do not change the evaluated version afterward.

See SUBMISSION.md for a ready-to-paste application description.