Integrations

MCP (Claude Desktop, Claude Code, Cursor, VS Code)

pip install "veramem[mcp]"
veramem mcp-config -u your-name     # prints the JSON to paste into your client

Tools exposed: remember, recall, ask, feedback, check_action, correct, forget, history. The acting user comes from VERAMEM_USER; the model can't pick another user's id.

REST API

VERAMEM_API_KEY=$(openssl rand -hex 24) veramem serve --host 0.0.0.0

The interactive OpenAPI docs are at /docs. Main endpoints:

Method	Path	Purpose
POST	/v1/memories	add messages
POST	/v1/search	search, returns context ready for a prompt
POST	/v1/answer	grounded answer with citations
POST	/v1/feedback	outcome reward for a trace
GET	/v1/memories/{id}/evidence	resolve evidence pointers
GET	/v1/memories/{id}/history	version chain and ledger hashes
POST	/v1/grants	grant or revoke user↔agent, agent↔resource, team membership
POST	/v1/check-action	authority gate
GET	/v1/verify	ledger verification
DELETE	/v1/users/{id}	right to be forgotten

Warning

The REST API trusts the user_id you send. Put it behind your application's own authentication; don't expose it directly to end users.

Docker

VERAMEM_API_KEY=... docker compose up -d     # VeraMem + Ollama
docker compose exec ollama ollama pull qwen3:8b
docker compose exec ollama ollama pull nomic-embed-text

Any OpenAI-compatible model server

export VERAMEM_LLM=openai
export VERAMEM_LLM_BASE_URL=http://localhost:30000/v1   # vLLM, SGLang, LM Studio, llama.cpp
export VERAMEM_LLM_MODEL=Qwen/Qwen3-32B
export OPENAI_API_KEY=unused