Code Agent CLI

noetic is an interactive coding agent that runs in your terminal. You drop a noetic.config.ts in your repo, run noetic, and chat with an LLM that has read/write/edit/bash/grep/find/ls/lsp tools, plus a sub-agent system, persistent sessions, and a background daemon for long-running task automation.

It is built on top of @noetic/core — every long-lived behavior (the autopilot loop, validator queue, worktree reaper, file-watcher reconciler) is a Step composition. You can extend it without forking by writing a plugin.

Install

npm i -g @noetic/cli
# or: bun add -g @noetic/cli

You also need an OpenRouter API key in OPENROUTER_API_KEY (or NOETIC_API_KEY).

First run

cd path/to/your/project
noetic

If no config file is found, the CLI uses CLI-arg defaults: model anthropic/claude-sonnet-4, max turns 50, and the OpenRouter key from your environment. The TUI launches and you can start chatting.

For anything beyond a one-shot trial, write a noetic.config.ts:

import type { AgentConfig } from '@noetic/cli';

export default {
  model: 'anthropic/claude-sonnet-4',
  apiKey: process.env.NOETIC_API_KEY!,
  cwd: process.cwd(),
  maxTurns: 100,
} satisfies AgentConfig;

See Configuration for the full schema.

Mental model

Three layers stack to make up a session:

1. The TUI (Ink + React) renders the chat, autocomplete, slash commands, and modals (ask-user, plan approval, model picker, session resume).
2. The harness is a AgentHarness from @noetic/core configured with a tool pool, memory layers, skills, and an OpenRouter LLM client. Each user turn drives one harness execution.
3. The daemon (optional) is a separate long-lived process that runs the task system — autopilot, validator, health, reconcile flows — alongside your interactive sessions.

What you can configure

Each surface has its own page:

• Configuration — noetic.config.ts schema, discovery order, env vars, built-in commands, built-in tools.
• AGENT.md — project & user instruction files surfaced into every turn.
• Rules — granular .agent/rules/*.md files attached to AGENT.md.
• Hooks — worktree lifecycle hooks (pre-start, post-start, post-merge, pre-remove).
• Skills — markdown + YAML personas, sub-agent presets, slash-command-invocable workflows.
• Plugins — npm-installable extensions that contribute tools, memory layers, skills, commands, footer UI, sub-agent presets, reminder triggers, and language servers.
• Custom flows — programmatic state machines composed from every / fork / spawn / branch / step.run, plus plan-mode JSON flows.
• Tasks — the unified noetic tasks system: hierarchies, kanban, autopilot, validator runs.