Goal-first recipes.

These are the first tasks Octopus should make easy: install it, run one local Goal, open the Goal/Need/Feed app, connect a model, and collect evidence.

Local App

Observe a real run

Start Octopus locally, open the app, and watch the current Goal, Need, Feed, and pixel pet state from the real state file.

octopus first-run "make this repo easier to use"
octopus start --open
Watch Goal, Need, and Feed update
Local Repo

Start the whole product in a repository

Run this from the project you want Octopus to inhabit. The user steers Goal; the agent handles Need, routing, Feed, memory, and harness signals.

octopus first-run "make this repo easier to use"
octopus start --open
octopus chat "make setup clearer"
Model

Use Codex login or an OpenAI-compatible model

Provider setup is runtime plumbing. It should make clean-brain and tentacle thinking available without leaking provider details into Need text.

codex login
export OCTOPUS_LLM_BACKEND=codex
export OCTOPUS_LLM_CODEX_COMMAND=codex
octopus provider status
octopus first-run --live "make this repo easier to use"
Evidence

Prove the product can run locally

Use evidence commands when preparing a release or diagnosing a machine. They stay observable surfaces; the first product path remains Goal input.

octopus start --check
octopus doctor
octopus report
octopus preflight

What each recipe should reveal

  • The brain emits Need without choosing an implementation.
  • A tentacle carries the local reasoning needed to turn one Need into useful Feed.
  • Heartbeat, memory, and harness evidence can improve future supply without polluting the main brain context.