Community documentation / 1.0.1rc4

Prepare for first recall.

Install from a verified source archive or, after publication, from PyPI. The release candidate API may change before stable.

01. Install the Community preview

Python 3.10 through 3.12 is supported. The MCP and model-export extras are explicit.

uv install
$ uv tool install --python 3.12 "lummenna[mcp,export]==1.0.1rc4"
Installs the lummenna and lummenna-mcp commands in one isolated environment.
ready

02. Initialize the palace

model provision
$ lummenna init --device generic
$ LUMMENNA_SOVEREIGN=false lummenna model download bge-small-en-v1.5
$ lummenna memory store "We chose SQLite for the cache layer" --room decisions
$ lummenna memory retrieve "cache layer"
ready

03. Run the application

serve
$ lummenna serve
Observatory: http://localhost:8848/
API reference: http://localhost:8848/docs
ready

04. Connect an MCP client

The MCP server exposes search, store, assemble, turn, feedback, status, and dashboard tools.

mcp configuration
{ "command": "lummenna-mcp" }
ready

05. Use the Python API

python
from lummenna.config import LummennaConfig
from lummenna.data.schema import get_connection
from lummenna.force.mnemonic.store import store_memory
ready

06. Prepare production deliberately

Use the committed uv lock, set an API key, use persistent store and model volumes, and keep the service on loopback unless network access is intentional.

secure start
$ export LUMMENNA_API_KEY=$(openssl rand -hex 32)
$ lummenna serve --host 127.0.0.1
ready

07. Build the React frontend

npm, pnpm, and Bun build the frontend only. Each supported path uses a committed integrity lockfile and disables dependency lifecycle scripts.

locked frontend installs
$ npm ci --ignore-scripts && npm run build
$ pnpm install --frozen-lockfile --ignore-scripts && pnpm run build
$ bun install --frozen-lockfile --ignore-scripts && bun run build
ready