lumid.data

A data management service with two surfaces under one URL:

• Traditional CRUD — raw SQL over Postgres (/sql) and S3-compatible object storage (/storage).
• LLM-driven data agent — /agent/v1 orchestrates the same CRUD endpoints plus deterministic schema-card and replay/materialization tools over an LLM tool-use loop; /mcp exposes the CRUD tools to any MCP client. Same auth, same audit log, no privileged backdoor.

Quick start

# Install
uv sync --all-packages --group dev

# App against an external Postgres + S3 (URLs from .env)
cp .env.example .env       # then edit values
docker compose up -d

# Or bring up the full bundled demo stack (Postgres + MinIO + Redpanda)
docker compose --profile bundled up -d

# ...or via the CLI
uv run lumid-data stack up

The service comes up at http://127.0.0.1:9100 (/docs for the OpenAPI browser). Set LUMID_DATA_LLM_API_KEY in .env to unlock /agent/v1.

# Direct CRUD (auth is a no-op by default; the Authorization header is
# optional unless an IdentityProvider plugin is registered).
curl -X POST http://127.0.0.1:9100/sql/v1 \
  -H "Content-Type: application/json" \
  -d '{"query": "SELECT count(*) FROM users"}'
curl -X PUT http://127.0.0.1:9100/storage/v1/object/lumid-data/hello.txt --data 'hello'

# Agent (streaming SSE)
curl http://127.0.0.1:9100/agent/v1 \
  -H "Content-Type: application/json" \
  -d '{"goal": "how many users do we have?", "skills": ["data_retrieval"]}'

CLI

lumid-data stack {up, down, restart, ps, logs, build, pull}
lumid-data sql query "SELECT 1"
lumid-data storage {get, put, ls, rm}
lumid-data agent run "..." --skill data_retrieval
lumid-data admin {audit, runs, run}

License

Apache License 2.0. See LICENSE.