Skip to main content

Quickstart

This quickstart gets Inferoa running against an OpenAI-compatible endpoint.

Install the CLI globally:

npm install -g inferoa@dev

@dev installs the newest Inferoa build published from main. The npm latest dist-tag is reserved for stable release versions. Node.js 24 or newer is required.

Then start the setup wizard and launch the TUI:

inferoa setup
inferoa

Pass a prompt as an argument to start a session and submit it as the first user turn:

inferoa "Inspect this repository and list the test entrypoints."

Run a one-shot prompt without opening the interactive interface:

inferoa --print "Inspect this repository and summarize the test entrypoints."

Basic Loop​

First Commands​

Slash commands live inside the TUI. See Slash commands for the full registry.

  • Use /setup to change provider, model, endpoint, web search, or Omni configuration.
  • Use /system (also /endpoints) to inspect model, web search, Omni, and runtime status.
  • Run /goal to start a recursive long-horizon goal.
  • Run /goal mode research for experiment-shaped work that needs repeated measurement.
  • Use /plan set for ambiguous work that needs an inspectable plan before execution.
  • Use /tokenmaxxing (also /cache, /rtk, /activity) to see token, cache, RTK, and routing pressure.
  • Use /access to change this workspace's file and tool access mode.

Repository Development​

When working from a source checkout, install dependencies, build, and link the local CLI:

npm install
npm run build
make dev-bin
inferoa setup
inferoa

Useful development commands:

npm test
make docs-preview
make docs-build
make dev-unlink

Configuration is stored under ~/.inferoa/. Endpoint keys are stored in the local vault; config files store key references. See the Configuration reference for the full schema and environment overrides.

Next Steps​

Once Inferoa is running, explore the long-horizon workflows:

  • First session walks through a complete task loop — describing work, using modes, inspecting evidence, and resuming later.
  • Goal mode is the loop-engineering surface for recursive long-horizon goals.
  • Plan mode turns ambiguous scope into an inspectable plan before execution.
  • Research goals under Goal mode run benchmark-style iteration with metrics and failure evidence.
  • Architecture explains the system model and how the harness, context layer, and inference layer connect.