Project management for teams and their coding agents

Your agent plans the work.Beck keeps the plan.

When you approve a plan in your coding agent, it lands in Beck as epics and tasks. You can follow the run as it happens, answer the agent’s questions from anywhere while it keeps working, and review what it shipped.

See plansWatch the loop

Free to start. Plans and Free are on the next page.

A demo of the loop. In the terminal, a coding agent plans sign-in with orgs in plan mode and the plan is approved. It calls record_plan, and the five tasks land in the Beck backlog. On the Now page the run goes on track, checks go from 38 of 42 to 42 of 42, and the first task goes to review. The agent asks whether to keep sessions in the auth provider, keeps going with that default, and the question waits on you until you keep the default. You accept the first task: one of five done.
  • Plans outlive compaction
  • Live on Now, with checks
  • Answer from your phone
  • Claude Code, Cursor, any MCP host

The problem

Three problems with agent work today.

  • The agent planned for twenty minutes. Then the conversation compacted and the plan was gone.
  • It made a call you would have made differently. You found out in the diff.
  • Someone asks what's shipping this week. The answer is in a terminal on your laptop.
A terminal holds the agent's plan. The conversation is compacted and the plan disappears from it. A hook tells the agent to call Beck's resume tool, which returns the task in progress and the plan document, plans/sign-in. The document in Beck never changed, and the agent picks up where it left off.

The conversation forgets. The plan in Beck doesn’t, and the agent resumes from it.

The Plan step: Your agent plans with the repo in front of it.

How the loop works

Plan. Record. Run. Ask. Review.

01 · Plan

Your agent plans with the repo in front of it.

In plan mode it reads your code and what Beck already knows: what's in progress, what's waiting for review. Then it proposes the breakdown, and you approve it where you already work.

02 · Record

One call, and the plan is somewhere your team can see it.

record_plan saves the plan as a document and creates the epic and its tasks in order, with acceptance criteria. It stays in Beck after the conversation ends or compacts.

03 · Run

Follow the run without the terminal.

Now shows what the agent is on, whether its checks pass, and what it's waiting on you for. If it goes quiet, Now says so.

04 · Ask

Questions come with a default.

When a choice is yours to make, the agent records the question and the default it's using, then keeps working. Your answer reaches it on its next call.

05 · Review

You accept the work, not a summary of it.

Finished tasks wait for review with what the agent tried, what passed and what didn't. Accept it, or reject it with a reason the agent reads when it picks the task back up.

The agent asks whether to keep sessions in the auth provider or its own database, and keeps writing code with the default. The question appears on the Waiting on you page on a phone. Someone taps Keep default. When the agent finishes its task, Beck tells it the default was kept.

Decisions

It asks you, and keeps going while you decide.

When a choice is yours, the agent asks with the default it’s using, then keeps working. The question waits for you on a page made for your phone.

  • Keep the default with one tap, or type what it should do instead.
  • The agent hears back on its next call, with the tasks your answer affects.
  • Costly to reverse? It says so, and can hold the affected tasks until you answer.

Setup

Works with the agent you already use.

Beck’s MCP server works in Claude Code, Cursor and any other MCP host, against https://beck.bot with a workspace token. Agents read the rules at /skill.md and /llms.txt.

In Claude Code, the plugin adds two hooks: it records the plan when you approve it, and resumes at the start of every session. Cursor has Beck in its marketplace.

  • Claude Code
  • Cursor
  • Codex
  • Windsurf
  • Grok
  • any MCP host
How working with an agent in Beck goes
install
# Any MCP host
npx @becklabs/beck-mcp-server
# BECK_API_TOKEN=beck_…  BECK_API_URL=https://beck.bot

# Claude Code: record-the-plan and resume hooks
/plugin marketplace add beckhq/beck-plugin
/plugin install beck@beck

Pricing

$12 if you’re shipping. Free if you’re looking.

Pro is you and your agents. Team is a shared backlog. Your own keys work on every plan.

Free

$0/month

Prove the loop on a personal project.

  • Unlimited workspaces
  • 50 AI credits / month
  • MCP, REST and the plugin
  • Your Anthropic or OpenAI key

Pro

$12/month

You and your agents, with more AI credits included.

  • 250 AI credits / month
  • Your own keys stay unlimited
  • Same MCP and REST
  • Cancel anytime

Team

For teams

$15/seat/month

Shared workspaces. One seat per teammate.

  • 250 AI credits / seat
  • Shared workspaces and roles
  • Invite seats as you grow
  • Everything in Pro

Credits are for Beck’s own AI, used by Plan it and spec help. Breaking down a spec costs 10, expanding it costs 5, a question costs 2. Daily limits apply when you use Beck’s AI (Free 5, Pro 50, Team 100). Your own keys skip credits and limits.

Fair questions

The objections, in your words.

I’ve tried agent workflows. They hallucinate scope.
Scope drifts when the plan only lives in a context window. In Beck the plan you approved is on the list, every task has acceptance criteria, and the agent asks before it changes direction. You reject with comments and it iterates on the same task.
Is this just another board with AI slop?
Beck doesn’t write your plan with its own AI. Your agent writes the plan; Beck records it, shows the run, and keeps the agent’s questions until you answer. If you don’t use an agent, the backlog and task review work like a traditional project management tool.
Will my team actually use another tool?
Start on Free, alone. Your team doesn’t need a terminal to see what the agent is doing. If the loop is real, invite seats. If it isn’t, you haven’t paid.

FAQ

Leftover questions.

Does Beck plan the work for my agent?

No. Your agent plans with the repo in front of it, usually in plan mode, and records the approved plan in Beck as epics and tasks. Beck keeps it across sessions, shows the run on the Now page, and holds the agent's questions until you answer.

What agents work with Beck?

Claude Code has a plugin that records the plan when you approve it and resumes each session. Anything else that speaks MCP (Cursor, Grok Bot, and others) runs npx @becklabs/beck-mcp-server. REST lives under https://beck.bot/api/v1 with a beck_ token.

Do I need my own AI keys?

No. Free includes 50 credits. You can add Anthropic or OpenAI keys on any plan, including Free, and skip Beck credits entirely.

Can I use Beck without agents?

Yes. The backlog, icebox, and review work for people. Teammates who are not coding can write Documents and use Plan it to turn a page into stories. Agents make the loop faster; they aren't required.

Is GitHub a real integration?

You can paste a branch name and PR URL onto an epic. There is no GitHub App, OAuth, or repo sync.

Record your agent’s next plan in Beck.

See plans

Free includes 50 AI credits a month.