TODOforAI CLI (tfa-cli)

Create and manage TODOforAI tasks from your terminal.

Install and run

Requires Node.js 20+.

npm install -g @todoforai/cli
tfa-cli "Fix the login bug"

todoforai-cli is an alias for tfa-cli.

Configuration

API URL resolution: --api-url flag → TODOFORAI_API_URL env → https://api.todofor.ai.

Auth resolution: --api-key flag → shared-device run token → shared credentials file → TODOFORAI_API_TOKEN env → device login.

Project, agent, and last-todo state are stored per API URL under per_api_url[<url>] in the config — switching between e.g. https://api.todofor.ai and http://localhost:4000 keeps each environment's defaults isolated. Legacy top-level fields are auto-migrated on first run.

Bridge

Installed automatically on Linux and macOS.

The CLI talks to the backend over WebSocket; shell execution, file I/O, and tool calls happen in the bridge running locally. On create/resume/template runs, tfa-cli starts a detached todoforai-bridge process if needed (the bridge enforces its own single-instance lock, logs at ~/.todoforai/bridge.log). If bridge credentials are missing, the CLI runs todoforai-bridge login in the foreground first so you can see and approve the device-login URL. The bridge keeps running after the CLI exits, so long-running tasks survive Ctrl+D.

Disable with --no-bridge if you manage the bridge yourself (e.g. systemd, separate terminal). --no-edge remains supported as a deprecated alias.

Usage

Create a todo from a prompt

tfa-cli -n "Quick task"                    # non-interactive (run and exit)
echo "content" | tfa-cli                   # pipe from stdin
tfa-cli --path /my/project "Fix bug"       # explicit workspace

Start from a registry template

tfa-cli --template alternativeto-listing                          # interactive input prompts
tfa-cli --template f5bot-monitoring-setup --input "monitoring_details=My Brand"  # with inputs
tfa-cli --template f5bot-monitoring-setup --no-watch --json       # create only

When inputs are missing, the CLI prompts interactively (unless -n).

Inspect a todo (read-only)

tfa-cli --inspect <todo-id>

Prints the full chat log: messages, tool calls (type, status, path/cmd), results, and errors. No logo, no interactive mode.

Resume / continue

tfa-cli -c                     # continue most recent todo
tfa-cli --resume <todo-id>     # resume specific todo

Notifications

tfa-cli inbox                           # the bell: newest across all tabs
tfa-cli inbox messages --unread         # one filter: needs-you | messages | activity
tfa-cli inbox seen                      # mark everything read
tfa-cli notify "Deploy done" "v2 is live" --href /t/<todo-id>   # note to YOUR user (Messages tab + phone push)
tfa-cli project members                                          # who can be notified
tfa-cli notify --to [email protected] "Review?" "PR is ready" --href /t/<todo-id>   # note to ONE project member

notify goes to your own user by default. --to <email> reaches one member of the current project (--project <id>, $TODOFORAI_PROJECT_ID or the default project); non-members are refused (404). It is shown as from you (or your agent). There is no broadcast / "all members" flag, so loop over project members to reach the whole team. --subject sets the dedupe key (the same text is delivered once), --href must be an in-app path. Caps: 10 new notes/day to yourself, 20/day to teammates.

IDE integration (ACP)

tfa-cli acp speaks the Agent Client Protocol over stdio, so any ACP-capable editor can drive your agent: prompts from the editor chat, file edits as native diffs, permission prompts inline, and shell commands running on your machine through the bridge. One ACP thread = one todo. To log in explicitly, run tfa-cli login; --isolated / --user-id are not supported in this mode.

Zed — ~/.config/zed/settings.json:

{
  "agent_servers": {
    "TODOforAI": { "type": "custom", "command": "tfa-cli", "args": ["acp"] }
  }
}

Then open the Agent panel → + → TODOforAI.

JetBrains (IntelliJ, PyCharm, WebStorm… 2025.2+) — AI Chat → ⋮ → Add Custom Agent, which opens ~/.jetbrains/acp.json:

{
  "agent_servers": {
    "TODOforAI": { "command": "tfa-cli", "args": ["acp"] }
  }
}

Pick TODOforAI from the agent dropdown in AI Chat.

VS Code — no built-in ACP client yet; install a community ACP client extension (e.g. strato-space.acp-plugin) and add the same agent_servers entry to settings.json.

No global install needed — "command": "npx", "args": ["-y", "@todoforai/cli", "acp"] works in every host above. --agent, --project and --api-url apply to acp as well (e.g. "args": ["acp", "--agent", "backend"]). If tfa-cli is not on the editor's PATH, use its absolute path (which tfa-cli). Your agent settings show up in the host's mode picker, so you can switch agents per session without touching the config.

All Options

--path <dir>                    Workspace path (default: cwd)
--project <id>                  Project ID
--agent, -a <name>              Agent name (partial match)
--api-url <url>                 API URL
--api-key <key>                 API key
--template, -t <id>             Start from a registry template
--input <key=value>             Template input (repeatable)
--inspect, -i <todo-id>         Print full chat log (read-only)
--resume, -r [todo-id]          Resume existing todo
--continue, -c                  Continue most recent todo
--non-interactive, -n           Run to completion and exit
--dangerously-skip-permissions  Auto-approve all blocks (CI/benchmarks)
--allow-all                     Set permissions to allow all tools (no approval needed)
--no-watch                      Create todo and exit
--no-bridge                     Do not auto-spawn bridge
--no-edge                       Deprecated alias for --no-bridge
--json                          Output as JSON
--safe                          Validate API key upfront
--debug, -d                     Debug output
--show-config                   Show config
--reset-config                  Reset config file
--help, -h                      Show this help