Skip to content
MAESTRO

Examples

MAESTRO ships a couple of ready-made chains you can validate, import, and run.

Bundled chains

Section titled “Bundled chains”
ExampleWhat it shows
weatherA weather agent wired to an MCP server (mcp_servers.toml + chain.json).
financierA financial-analysis chain (chain.json) — a multi-step reasoning agent over financial input.

Each lives under examples/ in the care repo with a README.md.

Run a chain file

Section titled “Run a chain file”

Any chain.json follows the same flow — validate, import, run:

Terminal window
care validate examples/financier/chain.json      # preflight
care import examples/financier/chain.json --apply # into Memory
care run <chain_id> --execute --task "Analyse Q3 results"

Or skip Memory and just preflight + export:

Terminal window
care run <chain_id> --export chain.py             # export to a runnable module

The end-to-end workflow

Section titled “The end-to-end workflow”

The canonical MAESTRO loop, start to finish:

  1. Generate — type a task in the chat surface (or care generate "<task>").
  2. Run — Ad-Hoc runs inline; Production saves to the Library.
  3. Re-run — reopen a saved chain from the Library and run it on new input.
  4. Evolve — care evolve <chain_id> --wait --accept (or the Evolution screen) to improve it, then promote the winner.

Recording a demo

Section titled “Recording a demo”

examples/asciicast/recording_script.md has a keystroke script for recording an asciicast of a MAESTRO session.

See also

Section titled “See also”