chasqui 0.3.0

Chasqui stack generator — `uvx chasqui new <name>`: zero to a running AI chat agent in one command

chasqui — the stack generator

uvx chasqui new mi-agente

Zero to a configured, locally-running AI chat agent in one command — the project generator of the Chasqui stack (pick your channels: WhatsApp, Telegram, embeddable web chat widget).

The wizard asks only what becomes a .env variable (LLM provider/model, embeddings + dimension, where your Postgres is, which channels to install, service ports — so several stacks coexist —, WhatsApp credentials (how to get them), the Telegram bot token, the web widget's allowed embed origins, default language, first admin, optional media bucket / handoff notifications / deploy params) — then downloads the services at a pinned release tag (degit-style, no git history), brands them, writes the .envs, generates the shared secrets, and provisions: uv sync + npm install + createdb + migrations + admin seed. Every step that can't run prints its manual command and the scaffold continues.

Nothing is locked in: everything the wizard wrote lives in the generated .envs and can be changed any time (the LLM applies on the next message). The only provision-time choice is EMBEDDING_DIM — baked into the schema on the first migrate.

Commands

uvx chasqui new <name>                  # the wizard
uvx chasqui new <name> --defaults       # non-interactive (CI-friendly placeholders)
uvx chasqui new <name> --skip-provision # write files only
uvx chasqui new <name> --ref main       # scaffold an unreleased stack ref
uvx chasqui new <name> --source ~/path  # copy a local stack checkout (dev)

chasqui generate module <name>          # scaffold a Tool Module (run inside a project)
chasqui generate module <name> --with-models --with-admin

Development

uv sync && uv run pytest
uv run chasqui new demo --defaults --skip-provision --source ~/path/to/chasqui

Design: ADR-005. Releases pin the stack tag (src/chasqui/stack.py): CLI vX.Y.Z scaffolds services vX.Y.Z.

License

Apache-2.0