TUI

Bento's board in your terminal, with agents in sandboxes on your machine or on the hosted server.

Install the CLI

Requires Node.js 22.19+, curl and tar on macOS or glibc Linux (x64, ARM64). On Windows, use WSL 2. Local agents need Docker running.

curl -fsSL https://usebento.ai/install.sh | sh

Add Bento to PATH as the installer says, reopen your shell, and check:

bento --version

Quit the TUI before updating. Updates keep your projects, settings, credentials and Docker data.

bento update
bento

Choose where agents run

Local board and local agents

Build the sandbox image, then run setup. Use /usr/local/lib/bento/sandbox if the installer put Bento there.

docker build -t bento-sandbox:dev "$HOME/.local/lib/bento/sandbox"
bento setup

Bento runs its API and PostgreSQL locally, with data in ~/.bento. Rebuild the image when an update asks.

Hosted board and server agents

Sign in, then open the hosted board:

bento login --server https://app.usebento.ai
bento setup --server https://app.usebento.ai
bento --server https://app.usebento.ai

Repositories and credentials are configured on the server; no Docker needed. Set BENTO_URL=https://app.usebento.ai to skip --server.

Hosted board and local agents

Shared board at usebento.ai, agents on your checkouts:

bento --server https://app.usebento.ai --run-agents local

The TUI runs agents while open. To run queued work without it, start a runner:

bento runner --server https://app.usebento.ai

Work waits while the runner is offline. Automatic pull requests need server agents, since the server cannot reach local worktrees.

Set up a project

Open Settings with , or bento setup. Tab moves between fields, Control+S saves, Escape cancels.

Connect repositories

Under Repositories, connect a Git checkout on the machine that runs the agents. For each one, set:

• Install dependencies: runs once per sandbox. Sandboxes have Git and agent CLIs but no language runtime, so install it here. Leave blank to let the agent work it out.
• Build and test: how the agent validates changes, such as pnpm test.

Configure agents

Under Agents, add an agent:

1. Harness: the coding tool, such as Claude Code, Codex or pi.
2. Model: what the harness runs.
3. Skill: standing instructions for every run: the stage's expected outcome, write-up and changes.

Save a key under Model provider keys (keys you save are yours; an owner or admin shares a key with the team from the Web UI), or for local agents share a CLI login under Local agent sign-ins (trusted repositories only). See Coding agents for model formats and mid-run behavior.

Build the pipeline

Under Pipeline, arrange the stages a card passes through. Each has an agent, requirements, and an optional pull request.

Start with manual gates: the card waits after each run so you can review, steer, approve or send it back. Switch a stage to automatic once its requirements are reliable. Editing an agent updates every stage that uses it. See Pipelines.

Run the first card

1. Press n, enter a title and description, and create the card.
2. Select it in Backlog and press a to start the first stage.
3. Press Enter to watch the conversation. Focus the reply box to guide the agent while it works.
4. Review the output, artifacts, and diff. Return to the board and press a to approve.
5. Continue through the pipeline until the card reaches Completed.

A successful run does not approve a manual stage.

Common shortcuts

Export agents and pipelines to reuse a setup in another project:

bento agents export agents.yaml
bento pipeline export pipeline.yaml
bento agents import agents.yaml --project "Another project"
bento pipeline import pipeline.yaml --project "Another project"

Run bento --help to see every command and option.