Codeify CLI is the official terminal coding agent for codeify.cc. It runs locally in your repository, edits files, executes approved commands, manages sessions, and talks to Codeify through the OpenAI Responses API.
Run the same command in PowerShell, Command Prompt, Terminal, Bash, or Zsh:
node -e "fetch('https://codeify.cc/install.cjs').then(r=>r.ok?r.text():Promise.reject(Error('Download failed: '+r.status))).then(s=>Function('require',s)(require))"
Node.js 22.19 or newer is required. On Windows, the installer downloads a source archive when Git is unavailable; Git remains required on macOS and Linux. Re-running the command updates an existing installation. Then run codeify from any project directory.
On the first interactive launch, Codeify offers two authentication choices:
- Continue with Codeify OAuth
- Enter a Codeify API key
Both credentials are saved under ~/.codeify/agent/auth.json and can be replaced later with /login or removed with /logout. CODEIFY_API_KEY is also supported for CI and one-off sessions.
Smart Model Usage keeps your selected model in charge while letting it delegate suitable work to cheaper or specialized models available through Codeify. This saves the main model's context and reasoning for architecture, difficult implementation decisions, security-sensitive work, and final review.
Enable it inside an interactive Codeify session:
/smart on
Use /smart to check its state, /smart off to disable it, or change Smart model usage at the top of /settings. It is disabled by default.
When enabled:
- The main agent can inspect the current Codeify model catalog, including price and capability details, and choose another model for a bounded task.
- Delegated models are fully agentic. They can inspect the repository, use approved tools, iterate across multiple turns, and return usage and changed-file summaries.
- Read-only tools are the default. Shell commands, file mutations, custom tools, and browser control must be explicitly granted by the main agent for that task.
- Codeify credentials remain inside the CLI harness. Neither the main model nor delegated models can read the API key or OAuth tokens.
- Delegation is bounded by tool, turn, output, concurrency, and wall-clock limits, and delegated agents cannot recursively delegate again.
- Vision work can be sent to a cheaper image-capable model. Browser tasks run in a fresh isolated headless session with a required domain allowlist, blocked downloads, no saved credentials, and network writes disabled unless explicitly authorized.
Smart delegation is especially useful for repository search, repetitive inspection, summarization, formatting, test triage, mechanical edits, and inexpensive vision tasks.
- OpenAI Responses API streaming with tool calls and reasoning levels
- Read, write, edit, search, and shell tools
- Session resume, fork, tree, export, and sharing workflows
- Slash commands, themes, skills, and prompt templates
- Configurable approval, project trust, model catalogs, and provider authentication
- Interactive terminal UI with Ctrl+C and Escape cancellation
npm install --ignore-scripts
npm run build:offline
npm run check
./test.shThe compiled launcher is used by default for fast startup; set CODEIFY_USE_SOURCE=1 when developing against TypeScript sources.
Useful commands:
codeify --help
codeify update
codeify --list-models
codeify --version
CODEIFY_TIMING=1 codeifyThe Codeify provider defaults to https://codeify.cc/v1 and gpt-5.6-sol. Override them with CODEIFY_BASE_URL and CODEIFY_MODEL when testing compatible gateways.
| Package | Description |
|---|---|
packages/coding-agent |
Interactive Codeify CLI |
packages/agent |
Agent runtime and session state |
packages/ai |
Multi-provider LLM and Responses API layer |
packages/tui |
Terminal UI and rendering primitives |
The source is public for transparency and internal use. Codeify maintains this repository privately and does not currently solicit contributions or expect to accept unsolicited pull requests.
Codeify does not sandbox filesystem, process, network, or credential access. Run it only in repositories and environments you trust. See SECURITY.md and packages/coding-agent/docs/containerization.md for boundaries and isolation options.
MIT. Portions of the implementation are derived from MIT-licensed open-source work by Mario Zechner and contributors.