Works inside Claude Code, Codex, and Cursor

Trust what your AI ships.

Let AI build more of your software. Forge’s workflows carry it through every step, prove it works, and bring you in when it matters.

Start free →
Free for individuals. Start on your own; bring your team later. · Install the plugin →

Same AI, same ticket. The workflow is the difference.

Without Forge
Claude Code v2.1.251
Sonnet 5.5 · Claude Max
D:\Development\Acme\web-shop
>implement PROJ-482
Called plugin:linear:linear 2 times
PROJ-482: ask for a shipping address only when the order needs one. 6 sub-tasks, approved plan.
Straight to code
Write(src/checkout/address-rule.ts)
⌊ Wrote 58 lines
Bash(script: swap the address check in 23 files)
Left 5 files alone — their tests expect the old behavior.
Grades its own work
Bash(npm test)
⌊ 1,205 of 1,206 pass · 1 pre-existing failure
PROJ-482 is implemented, all 6 sub-tasks.
– One gap: a label on the order summary
⚠ The same 2 bugs are on this branch
– Email: “Ships to undefined” on digital orders
– Shipping label queued with no address
Its own tests passed, so it never saw them.
>
It saidDone, all 6 sub-tasks
✗ You got2 bugs nobody knew about
✗ RecapNothing on what was checked
✗ CostUnknown per ticket
With Forge
Claude Code v2.1.251
Sonnet 5.5 · Claude Max
D:\Development\Acme\web-shop
>forge, implement PROJ-482
Skill(forge-shiptoday:forge-autopilot)
Implement a Feature · Step 1: Check Prior Work
Reads before it acts
PROJ-482 already has an approved plan and 6 sub-tasks. Building on it, not starting over.
Steps 2–8 on Haiku 5.5 · ticket, readiness, stories, discovery
Asks you when it matters
Step 9: Write the Code
Wave 1 plan: address rule, checkout form, 2 tests
⌊ Approve the Wave 1 plan? → Approve
Proves the test works
Broke the fix on purpose: the new test failed. Restored: it passes.
⌊ Wave 1 committed
Waves 2–4 the same way: plan, approve, build.
PR #214 opened · all 6 sub-tasks
Step 11: AI-Augmented PR Review
I wrote this, so a fresh agent reviews it.
Agent(Independent PR review) Opus 5.5 · fresh context
Fresh eyes on every change
Review: 2 high
– HIGH Email: “Ships to undefined” on digital orders
– HIGH Shipping label queued with no address
Both reproduce. My tests missed them.
Steps 13–14: Apply and Verify Review Fixes
– Both fixed; a new check fails on the old code
⌊ Reviewer re-checked: fixed, physical orders unchanged
Shows its work
Run complete: PROJ-482 is done
– 6 of 6 sub-tasks, 2 bugs fixed before merge
– Recap: plan, 4 approvals, review, and fixes
Shows what it cost
– Est. cost for PROJ-482: $6.40
– Haiku 5.5 to look up, Sonnet 5.5 to build, Opus 5.5 to review
>
It saidDone, all 6 sub-tasks
✓ You got2 bugs fixed before merge
✓ RecapEvery check and approval
✓ Cost$6.40, the right model per step
Inside the AI tools you already use
Works with the skills you already have
Every change checked, with proof to share
Who it’s for

For anyone
shipping software with AI.

On your own, on a product team, or in a company without engineers. Forge works the same way; what changes is where you start.

Solo builder

“I built the feature. What should I check before I ship?”

Launch without fear and change it without breaking it. Forge checks your work before you ship and shows what passed and what didn’t.

  • Say what done means
  • Forge runs the checks in your AI tool
  • See what was checked, and what wasn’t

Product or engineering team

“AI doubled our pull requests. Our reviewers didn’t double.”

Ship at AI speed and keep review and production calm. Changes reach review already checked, a person signs off where you choose, and you can show what each ticket cost.

  • Checks before review, not after
  • A sign-off where it matters
  • A recap of every run

Building without engineers

“Our ops team built a tool with AI. Can we run the business on it?”

Build the tools your team needs and ship them with confidence. Forge checks each change, asks before anything risky, and ends with a recap of what was checked.

  • Say what the tool should do
  • Checks on every change
  • A recap you can share with IT
Why Forge

AI changed how fast software gets written.
Checking it hasn’t caught up.

Review becomes the bottleneck.

AI multiplies pull requests, and every one still waits for a person.

5× median review time on teams with high AI adoption.1

With ForgeChanges reach review already tested, and a first review says what must be fixed before merge.

“Almost right” ships.

AI code looks right at a glance, and nothing stops it from shipping.

In 45% of test cases, AI-generated code contained an OWASP Top 10 flaw.2

With ForgeTests, the security pass, and review run as steps in your workflows, using the skills you already have.

Nobody can say who approved it.

What was decided and checked disappears with the chat.

34% of organizations with an incident in the past year couldn’t tell whether AI-generated code contributed to it.3

With ForgeA person signs off where you choose, and every run ends with a recap of what was checked and decided.

AI bills climb with no ceiling.

Fetching a ticket and designing a fix run on the same top model.

By 2028, Gartner expects AI coding costs to pass the average developer salary.4

With ForgeEach step uses the model it needs, and you see what every workflow and ticket cost, in Claude Code and Codex.

Built to fit, not replace
Your rules stay exactly as they are
CLAUDE.md, AGENTS.md, and .cursorrules keep governing how the AI writes code. Forge uses them inside its workflow steps and adds the shared context, the approval points, and a recap of each run.
Works with the tools you already use
Your AI coding tools, plus any client that supports MCP, the open Model Context Protocol. Switch tools anytime; your workflows stay in Forge.
Builds on the skills you already have
Forge surfaces the skills you already have at the right step, and lets admins mark the critical ones as required, so every run asks for them.
See how Forge works with skills →
Your AI tool handles the code
Your AI tool reads and edits code with its existing access. Forge guides the work using context your tool sends it. That context may include code excerpts and may be saved for future workflows.
Privacy details →
Workflows

Workflows for shipping software.
Every step, every time.

Forge’s built-in workflows cover the software lifecycle, from a bug report to a release. A workflow sets the order, makes sure the checks run, and calls on the skills you already have in your AI tool where they fit.

Skills in your AI tool bring the know-how.

Instructions, tools, and expertise that live in your AI tool. Your AI reads them and decides when to use them. A skill can hold a whole procedure.

Code reviewSecurity reviewTest generation

Forge workflows hold the process.

A workflow sets the steps, hands them to your AI tool one at a time, asks for the relevant skills, and pauses where a person signs off.

Step orderRun stateApprovalsRun recap
See how Forge works with skills →
How It Works

Brief. Check.
Decide. Recap.

Forge holds the workflow and gives your AI tool one step at a time. Four things happen on every run.

Your workflows · standards
your standards flow down
Forge
brief · check · decide · recap
workflow runsthe recap
Claude Code Codex Cursor

Set it up once. It applies in every tool below.

1
Brief — the AI starts with your context
Each step starts with your requirements, architecture, and standards, so nobody re-explains them.
2
Check — the right skills at the right step
Tests, lint, dependency audits, and review run as steps, so almost-right work is flagged before it ships.
3
Decide — people stay in control
A person signs off only where it matters, such as before a write, a merge, or a release. Everything else keeps moving.
4
Recap — every run ends with one
What was checked and who decided, summed up when the run ends. Reviews say what must be fixed before merge. Estimated cost shows in your dashboard.
From a sentence to a workflow

Say what you need.
Forge takes it from there.

You don’t need to know the workflows by name. Say what you need in your own words, at the start of a session or partway through one.

You sayforge, make team invitations expire after 7 days
When something needs to change
Make a change with a plan

Describe the change, with or without a ticket. Forge shows the steps and starts, asking before code changes.

  • A clear approach
  • The change
  • A recap of what ran
You sayforge, triage the comments on PROJ-412
When feedback arrives
Catch up on comments

Find what still needs attention across the ticket and the pull request.

  • Outstanding feedback
  • Ready to discuss or address
What Forge does with it
Runs a workflow you already have

Forge starts the right one from the workflows you have switched on. If your request is unclear, it asks you one question first.

See the built-in workflows →
Works out an approach for this task

Nothing fits? Forge works out the steps for this one task, shows you how it will approach it, and starts. You can stop it or change its steps at any time, and every step that changes something asks first.

Or build your own workflow →
When to bring Forge in

Use your AI tool the way you do today. Say forge when the result will ship or someone else will rely on it.

Reach for Forge
  • A feature or a bug fix that will ship
  • A pull request that needs a review before it merges
  • A release to plan or cut
  • Work that needs a sign-off, a recap, or a second reviewer
  • Anything your team should do the same way every time
Just ask your AI tool
  • Questions, explanations, and reading code
  • Quick edits you’ll check yourself, like a rename or a typo
  • Spikes and throwaway prototypes
  • Anything you’d be fine redoing

Started without it? Say forge when the work turns real, and it picks up from where you are.

Go deeper

Past your first run.
Where to look next.

  1. Follow one feature from idea to reviewed pull request Each person reaches for the right workflow, and every run ends with a recap of what was checked. Walkthrough →
  2. Run one shared process across your team Shared workflows with approval points, and a dashboard of the work. For teams →
  3. See what AI cost, per workflow and per ticket Pick a model tier for every step, so the top models go where they matter. Token Intelligence →
  4. Earn the trust, then let AI drive Start with sign-offs. Once a workflow holds up, switch on Let AI drive. AI makes every decision, and every check still runs. AI-driven work →

We build Forge with Forge. Every change to Forge is written by AI, proven by Forge, and signed off by a person, in Claude Code, Codex, and Cursor.

Get started

Install, then
start with a task.

Free for individuals. Set it up once, then ask for what you need.

  1. Step 1

    Set up Forge in your AI tool

    Claude Code, Codex, or Cursor. Install the plugin and sign in.

    Install guide →
  2. Step 2

    Ask Forge for a task

    Say “forge” and what you need, in your own words.

    forge, what did we just build?

  3. Step 3

    See what else you can ask

    Open Workflows in the Forge dashboard.

    See every workflow →
Bringing your team?
Invite them from the Forge dashboard.
Start free →
Pricing

Start free.
Bring your team when you’re ready.

Free for individuals, forever. Team features are free for up to 10 members during beta. Business follows when you need enterprise security and compliance.

Individual
Freeforever

Checks and a recap for your own AI code changes, inside the tools you already use.

What's included
✓Works with Claude Code, Codex, Cursor & any MCP tool
✓Built-in workflows
✓Workflow configuration
✓Your existing rules and skills
✓Model routing on supported tools
✓Basic analytics and audit history
Beta
Team
$0 per month Normally fifty dollars per month. Free while Forge is in beta.

Shared workflows and visibility for up to 10 members.

Everything in Free, plus
✓Up to 10 members
✓Shared workflows
✓Roles & permissions
✓Organization-wide analytics
✓Token Intelligence

Team features are free for up to 10 members during beta. Invite teammates from your dashboard when you’re ready. Team is $50 per month once beta ends. No commitment, cancel anytime.

Business
Coming Soon

For organizations with broader rollout and support needs.

Everything in Team, plus
✓Unlimited members
✓Enterprise SSO
✓Enhanced roles & permissions
✓Priority support
✓Planned compliance capabilities

Interested in early access? Talk to us →

Start free →

Forge runs inside your existing AI tools. Model routing and token measurement depend on the tool. Cost estimates use recorded tokens and published model rates; they may differ from your subscription bill.

Common questions

What to expect
before your first run.

A few details before your first workflow. Read all frequently asked questions →

Your AI tool reads and changes code using its existing access. Forge supplies workflow instructions; its service does not directly access your repository, CI/CD, or production. Your tool handles your keys; Forge doesn't ask for them. We do not use your data to train AI models. Context your tool sends may include code excerpts and may be saved for future workflows. Read the privacy details →
Workflows run in Claude Code, Codex, Cursor, and compatible MCP clients. The dedicated plugins add guardrails: file edits and writes to connected tools wait while a question or an approval is open. Model routing works in Claude Code and Codex. Token usage is measured in Claude Code and supported Codex versions, and not in Cursor. Compare the tools →
No. Triaging review comments, previewing a release, and recapping your work all run without one, and results are handed to you to paste. Workflows built on a ticket, such as breaking down a feature, say so before they start. Connect one through your AI tool when you want workflow results written to the ticket.
No. A Claude Code workflow splits one big job across many agents in a session, so it finishes faster. A Forge workflow is how a piece of work gets from ticket to merge: the steps in order, the checks that must run, where a person signs off, and the recap at the end. It works the same in Claude Code, Codex, and Cursor, and your team shares it. They solve different problems, and you can use both.
An AI code reviewer checks a proposed change. Forge also guides the work that leads to it: a plan, approval before code changes, a regression test that catches the bug, review, and a recap of the evidence. Keep the reviewer you already use; it can be one of those checks. See how they fit together →
Use them. Forge works across Claude Code, Codex, and Cursor, puts a sign-off where it matters, shows what runs cost in Claude Code and Codex, and ends each run with a recap of what happened.
No. AI output varies with the model, the context, and the task. Forge makes the process repeatable and shows the checks and decisions behind a result.
Today, yes. Forge is built for shipping software: code changes, reviews, and releases. It doesn’t check reports, spreadsheets, or other non-code work yet.
A skill teaches your AI how to do something, and the AI decides when to use it. A workflow is the process itself: Forge holds the steps, hands them over one at a time, stops where a person signs off, and ends with a recap. Workflows run your skills, so you keep both. See how Forge works with skills →
No. One person can install Forge, set it up, and run it. Team features are there when you add people.
Install the plugin in your AI tool, sign in, and ask Forge for what you need. If the workflow that fits is on, it runs on your real task. If it’s off, Forge offers to switch it on, and you say yes once. A team admin can switch on more from the Forge dashboard. See three ways to start →

Start with one task.
See what was checked.

Install Forge in your AI tool and start with one real task. Free for individuals; your tools, rules, and skills stay with you.

Start free → See how to install

We build Forge with Forge, in Claude Code, Codex, and Cursor.