@agentify/cli

The Agentify CLI pairs a local workspace with Agentify Chat.

Agentify Chat lets you send encrypted prompts from chat.agentify.sh to a coding agent running on your own computer. The server routes ciphertext only; the workspace key and runner tokens stay local.

Public Support

• Product: https://chat.agentify.sh
• Issues: https://github.com/agentify-sh/chat/issues
• Email: [email protected]

The CLI does not include an Agentify billing flow. Provider usage, subscriptions, and local CLI authentication remain controlled by the provider CLIs you run on your own machine.

Install

npm install -g @agentify/cli
agentify --help

The hosted setup flow at https://chat.agentify.sh can also install the CLI for a single user account without a global npm install.

Main Commands

agentify chat pair --pairing-ticket <ticket> --workspace "$PWD" --label "$(basename "$PWD")"
agentify chat daemon --runner codex
agentify chat daemon --runner claude
agentify chat daemon --runner gemini
agentify chat daemon --runner opencode
agentify chat daemon --runner claude --model haiku
agentify doctor

agentify chat pair registers this machine as the local decrypting device for a workspace. agentify chat daemon --runner codex runs Codex through structured JSON events for reliable encrypted output streaming. Claude, Gemini, and opencode are available through the same --runner flag.

For Claude Code, the built-in runner uses your local Claude Code subscription login in headless claude --print mode. If headless auth still fails, create a long-lived subscription token:

claude setup-token
export CLAUDE_CODE_OAUTH_TOKEN="<token printed by claude setup-token>"
agentify chat daemon --runner claude

The built-in Claude runner ignores inherited ANTHROPIC_API_KEY, CLAUDE_API_KEY, and ANTHROPIC_AUTH_TOKEN so stale API keys do not override your Claude Code subscription login. Use agentify chat daemon --runner-command ... for custom local CLI commands or explicit API-key workflows.

By default, Agentify starts in fresh structured-run mode: every browser message launches a new local CLI invocation with that CLI's default model. In this mode, /model is informational only and does not change future messages. Pass --model <model> to pin the model for fresh runs, for example --runner codex --model gpt-5.5, --runner claude --model haiku, --runner gemini --model gemini-2.5-flash, or --runner opencode --model opencode/qwen3.6-plus-free.

Persistent live mode is advanced and experimental. Start it explicitly with --persistent or --persistent-experimental only when you want Agentify to keep the selected local CLI open between turns. In persistent mode, Agentify parses the virtual terminal screen, filters control output before sending browser chunks, and attempts vendor-specific /model ... automation. If model switching cannot be verified, use the local CLI picker directly or restart the fresh runner with --model <model>.

Security Model

• Prompt and result payloads are encrypted before they leave the browser or local runner.
• The server does not receive the workspace encryption key.
• Poll tokens stay on the local machine.
• Browser enqueue tokens and approval state stay in the browser.
• V1 direct mode gives remote prompts the same local access as the selected CLI would have if you typed the prompt on this computer.
• Local logs stay on the paired computer.

Local Development

To exercise the npm wrapper against a local Rust build:

AGENTIFY_CLI_BIN=/absolute/path/to/agentify node ./bin/agentify.js --help

To stage the current Rust binary into the npm package layout:

cargo build --release
npm run stage:bin -- --force
node ./bin/agentify.js --help

To verify release readiness:

npm run release:check

Release automation

For one-command release prep across all supported platforms:

npm run release:all

From the workspace root, you can run the same flow with:

bash ../scripts/release-agentify-cli.sh

The release pipeline is non-publishing by default and runs:

• cross-platform Rust builds
• multi-target npm wrapper staging
• local version/release checks
• package verification
• release payload leak scan

Use explicit flags to publish:

npm run release:all:publish -- --npm-tag latest

or:

bash ./scripts/release-all.sh --publish-npm --publish-git --auto-commit --create-tag --push-git

Useful options:

--targets darwin-arm64,linux-x64
--publish-npm
--publish-git
--auto-commit
--create-tag
--push-git
--skip-leak-scan
--skip-cargo-test
--skip-npm-tests
--dry-run

Identity checks are enforced on publish:

• npm account must be agentify
• GitHub account must be waml

During local incubation, when only some platform binaries are staged, use:

npm run release:check -- --allow-missing-targets

Do not publish until the packed artifact has been inspected and the release check passes.