Skip to content
Bento
GitHubSign up
See what’s new

For engineers who

An agent graph that orchestrates coding agents from idea to pull request. Track all your open features while keeping the context.

Start building for freeDownload for Mac
Or install the CLI
curl -fsSL https://usebento.ai/install.sh | sh
View install script
A feature block moves left to right through planning, implementation, and a pull request.

Works with your favourite harnesses and models

Claude CodeCodexCursorAntigravityDeepSeekOpenCodePiPoolside

Bring agents into your existing software development lifecycle

Coordinate agents across many features at once. Your team shares one board, so progress and context stay visible.

Keep track of all your coding agent sessions

Every engineer has multiple coding agent sessions going, and it is easy to lose track and sight of them. One board keeps every sessions’ status trackable

Team boardVisible to the team
  1. 01Product investigation
    Checkout recoveryagent working
  2. 02UI/UX design
    Billing empty statesin review
  3. 03Engineering requirements
    Usage-based billingnot started
  4. 04Implementation
    Move audit log off hot pathagent working
  5. 05Code review
    Rate limit the public APIwaiting at gate
  6. 06Quality engineering
    Retry timed-out webhooksdone

Build your process into a pipeline

You already have a process you want the agent to follow: product investigation, design, spec, implementation, code review, QA. Today you have to prompt each step by hand.

Stage
PM
Agent
Product manager
Model
Claude Sonnet 5.0
Skill
Define the user problem, scope, and acceptance criteria. Produce a product brief the team can build from.

Sessions shared and reviewable by the whole team

Run your agents in isolated sandboxes. Pick up and resume work from any device and share context with your team

The context goes with the code.

Each feature keeps its own plans, designs, and review artifacts. Stage write-ups are committed alongside the code, so the next agent starts with what the last one learned.

Artifacts by featureExample
  • Team invitations3 artifacts
    • product-investigation.md
    • ui-ux-design.md
    • qa-report.md
  • Activity exports3 artifacts
    • product-investigation.md
    • ui-ux-design.md
    • qa-report.md

Control how agents move.
Keep the boundaries clear.

Choose how each stage clears, then run every feature in an isolated environment with access limited to what it needs.

Manual when judgment matters. Automatic when requirements pass.

Set the gate for each stage. Review uncertain work yourself, or let clear requirements move routine work forward.

Manual approvalhuman decision

Review the diff, approve it, or send the work back.

Agent run finishedApprove
Auto approvalrequirements

The card advances only when every requirement passes.

  • Tests pass
  • Artifacts committed
  • Review complete
Every decision stays in the feature history

One feature. One isolated workspace.

Each card gets its own worktree and sandbox, so concurrent agents don’t interfere with your machine or one another.

feature cardCheckout recoveryfeature/checkout-recovery
  1. sandbox 01running
    checkout/webisolated worktree
  2. sandbox 02running
    checkout/apiisolated worktree

Credentials stay outside the sandbox.

Model API keys stay encrypted in Bento: agents get a proxy token for each agent session. Encrypted personal and shared credentials and short-lived GitHub tokens are supplied only when required.

How credentials are handled
trusted service
Model keyproxy tokenagent session
GitHub tokencheckout/web14m 32s
Organization secretruntime onlyhidden
not stored in the agent workspace
Integrations

Start a card from Linear or Slack.

Kick off a feature in the tools your team already uses, then let Bento run the pipeline.

Linear

Tasks created in Linear can automatically create a feature card in Bento and start the pipeline.

  1. A Linear task is created
  2. Bento opens a feature card
  3. The pipeline starts
Connect your account in BentoSign up to connect Linear

Slack

Message @bento to create a card and start the pipeline. Approve in Slack and see agent output.

  1. Message @bento to create a card
  2. The pipeline starts
  3. Approve the gate in Slack
  4. Read the agent output in the thread
One-click installAdd to Slack
FAQ

Questions?

What to know before you run Bento.

Is Bento open source?

Yes. You’re free to run the Bento TUI, Mac app, and Web UI locally, or self-host Bento Server for your team. If you’d rather use a managed service, usebento.ai includes a free plan and paid plans for higher usage.

Does Bento support BYOK?

Yes. Bring your own API keys from your favourite model providers. Bento does not offer model tokens with its free or paid plans. Your keys stay encrypted in Bento: agents get a short-lived proxy token instead, see Security.

Which platforms does Bento support?

Bento is available as a CLI/TUI for your terminal, a web application in your browser, and a Mac application for Apple silicon and Intel Macs.

Why use Bento over running agents directly?

Bento organizes your sessions without losing track of each feature’s status or context. Give each agent and pipeline stage a SKILL.md, and you can reuse the same instructions instead of repeating prompts.