Agent Memory Challenge submission sheet

Use these values when creating the evaluation request. Keep all issued keys out of this repository and out of free-text fields unless a field explicitly requests one.

Field	Declared value
System name	ReFind
Version	v0.1.0
Evaluation type	Textual Memory
Division	Academic Methods
Participation route	Code — platform deployment
Public repository	https://github.com/imlrz/ReFind
License	MIT
API authentication	None inside the isolated deployment
Add endpoint	POST /v1/memories/add
Search endpoint	POST /v1/memories/search
Health endpoint	GET /health
API entrypoint	app.main:app
Container port	8000
Planner model	openai/gpt-4o-mini through OpenRouter
Suggested Add concurrency	16
Suggested Search concurrency	16
Supported platform top_k	1–1000; 100 is recommended for the official run

The contact email and non-public credentials should be entered only in the organizer's application form, not committed here.

Platform build and startup

docker build -t refind:v0.1.0 .
docker run --rm \
  -p 8000:8000 \
  -e RETRIEVAL_MODE=agent \
  -e LLM_API_KEY="$OPENROUTER_API_KEY" \
  -e LLM_MODEL=openai/gpt-4o-mini \
  -v refind-data:/data \
  refind:v0.1.0

No evaluation or memory-system key is embedded in the image. The maintainer must inject an OpenRouter-compatible GPT-4o-mini key at runtime because the declared retrieval method uses that model during Search.

Pre-submission evidence

• Unit and contract suite: python -m pytest
• Local contract smoke: RETRIEVAL_MODE=bm25 python scripts/smoke_test.py http://127.0.0.1:8000
• Container health: Docker HEALTHCHECK calls /health
• Method and adaptation disclosure: METHOD_CARD.md
• API contract and configuration: ../README.md

Before starting Full, record the immutable Git commit and confirm that it matches the v0.1.0 tag. Full evaluation freezes this version under the challenge rules.