Start free
  1. Goal
  2. Signals
  3. Roadmap
  4. Plan
  5. Approve plan
  6. Build
  7. Approve pull request
  8. On call
  9. Cost

The agentic operating system for your whole company

Leadership sets the goals, product picks the work, agents build it, operations keeps it running and finance sees what it cost. It all runs in one workspace, on the models you choose, and a person approves what matters.

Start free Follow one goal

Bring your own model keys. Start with a free 10M-token trial.

Delivery

A program with 1 of 8 work items done, 1 building and 2 held for a person.A program with 1 of 8 work items done, 1 building and 2 held for a person.

Leadership

Objectives under the Merchant growth pillar, each with its key results.Objectives under the Merchant growth pillar, each with its key results.

Engineering

A crew’s reviewers: Architect and Security pass, QA advisory.A crew’s reviewers: Architect and Security pass, QA advisory.

Operations

The Payments on-call rota: Priya Nair primary and Alex Morgan secondary.The Payments on-call rota: Priya Nair primary and Alex Morgan secondary.

How it works

Follow one goal from the board deck to production

Every team works the same goal in one workspace. Agents do the work at each step, and a person decides at the two that matter.

  1. Leadership

    Set the goal

    Make instant payouts the reason merchants switch. Its key results say how you will know.

    Objectives with their key results and linked initiatives.Objectives with their key results and linked initiatives.

  2. Product and support

    Hear what customers ask for

    Asks from customers and sales are kept as signals. Bug reports from Slack, email, Zendesk and Intercom go to the triage queue instead.

    Signals from a client, a growth analysis and auditors.Signals from a client, a growth analysis and auditors.

  3. Product

    Put it on the roadmap

    Score it with RICE and place it under Now. The forecast comes from the people you actually have.

    A capacity forecast: 50% likely in 6 weeks, 85% in 15 and 95% in 23.A capacity forecast: 50% likely in 6 weeks, 85% in 15 and 95% in 23.

  4. Agent

    The Architect writes the plan

    It reads the spec and the code, writes the solution design and splits the work into items, each naming its repository.

    A program past scope and plan, waiting at the approve step.A program past scope and plan, waiting at the approve step.

  5. You

    Approve the plan

    Nothing is built until a person approves. Send it back with a note and the Architect revises it.

    You decide
    A plan approval for SEPA instant payouts with Approve, Refine and Reject.A plan approval for SEPA instant payouts with Approve, Refine and Reject.

  6. Engineering

    Crews build each item

    The Engineer writes the change and runs your tests. Security and QA review the diff. Watch it work, and step in from the same thread.

    A crew’s reviewers: Architect and Security pass, QA advisory.A crew’s reviewers: Architect and Security pass, QA advisory.

  7. You

    Approve the pull request

    Read the diff and the reviews, then approve, ask for changes or reject. Anything the checks could not verify is flagged.

    You decide
    A release decision held for a person because tests, build and lint were not verified.A release decision held for a person because tests, build and lint were not verified.

  8. Operations

    Keep it running

    The team’s rota covers it. When a Datadog monitor fires, the Oracle reads the code and logs and posts what it found.

    The Payments rota: Priya Nair primary and Alex Morgan secondary.The Payments rota: Priya Nair primary and Alex Morgan secondary.

  9. Finance and leadership

    See what it cost and what moved

    Spend is broken down by model and initiative, and hours are booked to the same initiative. The key result takes its next reading, entered by a person or recorded by a workflow.

    A timesheet booked to an initiative linked to a key result.A timesheet booked to an initiative linked to a key result.

Every team

What it does for each team

The same programs, people and approvals, seen the way each team works.

Leadership

See whether the strategy is moving.

  • OKRs
  • Delivery health
  • Weekly brief
  • Reports
Two objectives with their key results and linked programs.Two objectives with their key results and linked programs.

Product

Decide what to build next.

  • Signals
  • RICE
  • Roadmap
  • Capacity
  • Customers

Engineering

Ship with agents, keep the review.

  • Crews
  • Code Studio
  • PR review
  • GitHub and GitLab
  • Skills and memory

Operations

Keep production up.

  • On-call
  • Paging
  • Alert investigation
  • SLAs
The Payments rota: Priya Nair primary and Alex Morgan secondary.The Payments rota: Priya Nair primary and Alex Morgan secondary.

Support

Turn reports into complete tickets.

  • Slack intake
  • Email
  • Zendesk
  • Intercom

Finance

Know what agents cost, and where hours went.

  • Spend by model
  • Cost alerts
  • Timesheets
A timesheet filled from calendar rules, booked to an initiative linked to a key result.A timesheet filled from calendar rules, booked to an initiative linked to a key result.

Marketing and people

Run work that has no code.

  • Programs
  • Requests
  • Brand voice

Knowledge

Ask the workspace anything.

  • Oracle
  • Pages
  • Confluence import
  • Google Meet

Security

Pass the security review.

  • Audit trail
  • SSO
  • IP allowlists
  • Customer-managed keys

Any model

Use any model, and change it when you like

  • A model per role. Give the Architect one model and the Engineer another.
  • Your own server. Point AnyForge at Ollama or vLLM, or your Baseten account.
  • No markup. Your provider bills model use directly. People can also use their own Claude Pro or Max plan.
  • Anthropic
  • OpenAI
  • Google Gemini
  • AWS Bedrock
  • Vertex AI
  • DeepSeek
  • Meta
  • OpenRouter
  • Vercel AI Gateway
  • Ollama
  • vLLM
  • Baseten
  • Claude Pro or Max

Switch provider and the crews, reviews and records carry on as before.

Not another harness

Keep Claude Code. AnyForge governs the work around it.

AnyForge doesn’t replace your coding agent or its plugins. It’s the governance layer above them, so a larger team can run more agents and still see everything they do.

  • Keep your setup. Claude Code, Cursor, VS Code, Cline or Continue, with the plugins and skills you already use.
  • One plan and one approval path. Each agent reads its work item, updates it and asks a person for review.
  • See every call. Route model calls through AnyForge Control to see what each one did and what it cost.
# Connect Claude Code to AnyForge
$ claude mcp add --transport http --scope user \
    anyforge https://crew.anyforge.ai/api/mcp

Autopilot

Autopilot, once you trust it

Turn it on for a program, an initiative or one work item. It starts crews on approved work, sends failing CI back for a fix, and merges a pull request once every check passes.

Plan approval always waits for a person. Autopilot also hands the pull request back to you when:

  • Back to you

    A reviewer finds a serious issue

  • Back to you

    Coverage drops

  • Back to you

    The diff is over your size limit

  • Back to you

    A file is outside your allowed paths

Questions

Questions teams ask first

Does AnyForge replace Claude Code?

No. AnyForge is not a coding harness. Your developers keep Claude Code, Cursor and their plugins. AnyForge is the governance layer above them: one plan, approvals, an audit trail and the cost of every model call routed through it, so a larger team can scale agents and see what each one does.

Which models can I use?

Anthropic, OpenAI, Google, AWS Bedrock and Vertex, DeepSeek, hundreds more through OpenRouter, or your own server running Ollama or vLLM. You can pick a different model for each agent role.

Does anything merge without a person?

Not by default. Every plan and every pull request waits for a person. Once you trust the setup, Autopilot can merge pull requests that pass every check, and it hands anything else back to you.

Where does my code go?

Crews work on a temporary copy in an isolated sandbox, and model calls go to the provider whose key you added. On an annual contract, crews can run on your own infrastructure.

Do we need repositories to use it?

No. Programs that people deliver by hand get the same plans, dates and reports, so marketing, people and operations teams can run their work in AnyForge too.

How is it priced?

A flat platform fee per million tokens on your own model keys, with no seats and no subscription to use the platform. Every workspace starts with a free 10M-token trial.

Start with one team. Add the rest when you are ready.

No seats, and a free trial to start. Bring the model keys you already pay for.

Start free See pricing