Skip to content

README.md

Latest commit

 

History

History
91 lines (65 loc) · 2.43 KB

README.md

File metadata and controls

91 lines (65 loc) · 2.43 KB

Tool-Based Memory Example

This example keeps memory explicit. A NAT react_agent decides when to call get_memory and add_memory; Redis Agent Memory stores and searches the long-term memories behind those tools.

Install

This workflow uses NAT's react_agent, so it needs the langchain-backed NAT agent package in addition to this plugin.

For a published install:

pip install "nvidia-nat[langchain]" "nemo-agent-toolkit-redis"

For local development from this repository:

uv sync --group dev --extra test

Configure

cp examples/tool_based_memory/.env.example examples/tool_based_memory/.env

Edit examples/tool_based_memory/.env:

  • OPENAI_API_KEY for NAT's OpenAI LLM and Redis Agent Memory extraction
  • REDIS_AGENT_MEMORY_URL and REDIS_AGENT_MEMORY_NAMESPACE for the memory server
  • HOST_REDIS_PORT and HOST_REDIS_AGENT_MEMORY_PORT if the default local ports are already occupied
  • REDIS_STACK_IMAGE and AGENT_MEMORY_SERVER_IMAGE if you need to override the tested image tags

Start Services

Compose starts Redis Stack and Agent Memory Server for local development. Both ports bind to 127.0.0.1, and AMS auth is disabled.

docker compose \
  --env-file examples/tool_based_memory/.env \
  -f examples/tool_based_memory/compose.yml \
  up -d

Validate And Run

The runner loads examples/tool_based_memory/.env automatically and passes a stable user_id plus conversation_id into NAT so the same user can retrieve what was stored earlier.

uv run nat validate --config_file examples/tool_based_memory/configs/config.yml
uv run python examples/tool_based_memory/run_agent.py

Default identity:

  • user_id=demo-user
  • conversation_id=demo-session

Example custom run:

uv run python examples/tool_based_memory/run_agent.py \
  --user-id alice \
  --conversation-id alice-session \
  --input "Remember that I prefer concise answers." \
  --input "How should you answer me?"

If you prefer the raw NAT CLI, you can still run nat run directly against the same config. The runner is just the easiest way to provide stable session context locally.

Stop Services

docker compose \
  --env-file examples/tool_based_memory/.env \
  -f examples/tool_based_memory/compose.yml \
  down -v

Notes

  • Tool-based memory depends on the LLM choosing the memory tools.
  • Use agent_auto_memory when every turn should be captured and hydrated automatically.