Agent-Evolving-Library

Evolving agents through feedback-driven memory consolidation.

🔗 Access the HLE data here: Google Drive - HLE Data

How to run the code:

python experiments/run_experiment.py \
  --env    experiments/configs/envs/hle.yaml \
  --solver experiments/configs/solver/single_agent.yaml \
  --tool   experiments/configs/tool/default.yaml \
  --memory experiments/configs/memory/empty.yaml

Run SearXNG (for tool search provider)

experiments/configs/tool/default.yaml uses web_search_provider: searxng, so start SearXNG first:

cd deploy/searxng
docker compose up -d

Quick check:

curl "http://127.0.0.1:8080/search?q=OpenAI&format=json" | head

If results is empty and unresponsive_engines shows timeouts, SearXNG is up but the machine cannot reach upstream search engines. In that case, set proxy env vars before startup (SEARXNG_HTTP_PROXY/SEARXNG_HTTPS_PROXY/SEARXNG_ALL_PROXY/SEARXNG_NO_PROXY), then restart compose.

Stop:

docker compose down