Reka CLI

Chat with Reka models, generate images and video, and launch coding clients using your Reka account.

To try the current source instead of the published package, run npm ci, npm run build, and node dist/index.js from this repository.

Install

npm install -g @reka-ai/cli
reka

Requires Node 22.12 or newer. Use npx @reka-ai/cli to run without installing. Standalone macOS and Linux executables are also available with each release.

Run reka for a short starting guide, reka --help for all commands, or reka <command> --help for that command's options.

Connect your account

reka login
reka doctor
reka models

Login opens developer.reka.ai and saves an API key in ~/.reka/config.json. Use reka login --manual over SSH. Creating a key requires a payment method on your account.

For CI and headless environments, create a key at developer.reka.ai/keys and run reka setup --api-key <key>, or supply REKA_API_KEY through your environment or secret manager.

reka logout removes the saved key and prints the link to revoke it. It does not revoke the key on the server or clear an environment variable.

Chat

Choose a model ID from reka models and replace MODEL_ID below:

reka chat MODEL_ID --prompt "Explain this error"
git diff | reka chat MODEL_ID -p "Review this diff"
reka setup --default-model MODEL_ID
reka "Explain this error"

Replies stream in a terminal. Use --no-stream to wait for the full reply, or --json for the provider's response object. Piped input is appended to --prompt.

Launch a coding client

Install OpenCode, Claude Code, or Pi, then choose its model once:

reka setup opencode
reka launch opencode

Replace opencode with claude or pi for those clients. Pass client arguments after --:

reka launch opencode -- run "Review this project"
reka launch claude -- -p "Review this project"
reka launch pi -- -p "Review this project"

The client runs in your current directory with your terminal and Reka credentials. Its existing config files are preserved. Launching a client can make billable requests.

Use --model MODEL_ID to override the saved choice for one launch. reka launch opencode --config prints the configuration without a key, network request, or file write. Launching requires macOS or Linux, including WSL.

Generate media

Discover an endpoint and inspect its schema first:

reka models --capability text-to-image --json
reka schema MODEL_ID/text-to-image --json
reka run MODEL_ID/text-to-image --prompt "A red bicycle" --download

Use an endpoint ID returned by reka models. Supported inputs vary by endpoint. run waits for completion by default. Use --async to return the request ID immediately, then inspect or download the result:

reka status REQUEST_ID
reka result REQUEST_ID --download
reka cancel REQUEST_ID

Output URLs expire. Download files you want to keep. For long jobs, use --timeout SECONDS; exit code 6 means the job may still be running.

Scripts and agents

Supply REKA_API_KEY through your environment or secret manager. Avoid putting keys directly in command arguments, which can appear in shell history.

reka models --json
reka --help --json
reka chat --help --json
reka snippet MODEL_ID --lang python

Commands print JSON when output is piped or --json is set. In JSON mode, errors go to stderr as an error object with code and message, plus hint and request_id when available. launch passes through the client's own output and exit status. mcp uses protocol stdio.

To use the MCP server with Claude Code:

claude mcp add reka -- reka mcp

reka init installs the bundled agent skill in the current project's .claude/skills/reka-cli/ directory.

Reference

Command and configuration reference · HTTP contract · Changelog · Security policy · Reka docs · Support

Development

Use Node 24 and npm with the committed lockfile:

npm ci
npm run check

The checks run typechecking, unit tests, the build, and isolated npm/npx installation tests. They use local fixtures and need no API key. Live tests are separate and can submit paid jobs.

MIT licensed.