CrowCode CLI

  ▄▄▄▄▄ ▗▖
 ▟█▛███▄█▘   CrowCode
 ▜███████▙   agentic coding in your terminal
   ▐▌ ▐▌

CrowCode is Orchestree's agentic coding tool that lives in your terminal. Give it a task in plain English and it reads your files, edits code, runs shell commands, and verifies its own work — with every destructive action gated behind an approval prompt.

❯ crowcode "audit and improve test coverage"

● read_file(package.json)
  ⎿  1  {
● grep(/describe\(/)
  ⎿  src/loop.test.ts:12: describe('runAgentTurn', () => {
● bash(npx vitest run --coverage)
  ⎿  Test Files  3 passed (3)
● Coverage is at 84%. The untested paths are the abort handling in loop.ts …

Get started

Note: CrowCode CLI is not yet published to a public registry. It is built and run from an Orchestree monorepo checkout. A public installer (curl … | bash, Homebrew, npm) will come with the public release.

From the root of an orchestree checkout:

pnpm install
pnpm --filter @orchestree/crowcode-cli build

Then make the crowcode command available on your PATH (one time):

cd apps/crowcode-cli && npm link

Navigate to any project directory and run crowcode.

Requirements:

• Node.js 20+
• An Orchestree API deployment that includes the CrowCode agent endpoint (POST /api/crowgent/agent). The CLI defaults to https://api.orchestree.ai and tells you honestly if the deployment you point it at doesn't support the CLI yet.

No API key is stored on your machine — model access is brokered by the Orchestree API.

Usage

crowcode                          # interactive REPL in the current directory
crowcode "fix the failing test"   # one-shot task, then exit
crowcode -m max --resume          # pick a tier, resume the last session here
crowcode --yolo                   # skip approval prompts (writes + shell run freely)
crowcode --api https://api.orchestree.ai   # point at a different deployment

Model tiers: fast, pro (default), max — white-labeled as CrowCode Fast / Pro / Max. The underlying model slugs never leave the server.

Slash commands (REPL)

| Command | What it does | |---|---| | /model [fast\|pro\|max] | Show or switch the model tier | | /new | Start a fresh conversation | | /tools | List available local tools | | /config | Show current configuration | | /yolo | Toggle auto-approval of writes + shell commands | | /clear | Clear the screen | | /help | Help | | /exit | Leave CrowCode |

Local tools

CrowCode runs its tools on your machine — the server only decides what to do, your terminal decides whether to allow it:

• read_file, list_dir, glob, grep — always allowed (read-only)
• write_file, edit_file, bash — approval-gated: [y]es / [n]o / [a]lways this tool

Press ctrl+c during a turn to interrupt it; press it again at the prompt to exit.

Configuration & sessions

Config and session history live in ~/.crowcode (override with CROWCODE_HOME):

• config.json — API base, default tier, auto-approve preference
• sessions/ — per-directory conversation history; --resume picks up the latest session for the current directory

Reporting bugs

CrowCode CLI is early. File issues in the Orchestree repository, or flag them in your Orchestree workspace.

Data handling

• Your prompts and conversation history are sent to the configured Orchestree API, which relays them to model providers to generate responses.
• Tool execution (file reads/writes, shell commands) happens locally; only tool output is sent back to the model as conversation context.
• Session transcripts are stored locally in ~/.crowcode/sessions and never uploaded.