What Forge is, and who it’s for
Forge lets AI write more of your code and lets you trust what it ships. It checks each change, asks you when a decision matters, and keeps a record of what was checked and who approved it. Forge is a plugin for Claude Code, Codex, and Cursor. With Jira, Linear, or Notion connected, results are written to the ticket.
Your rules guide the code. Forge guides the steps. Your AI tools already support instructions and skills. Forge brings those into a repeatable workflow: brief the AI on the work, run the relevant checks, pause at the approval points you set, and record the results. A team can share that process across tools and people.
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.
No. One person can install Forge, set it up, and run it. Team features are there when you add people.
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.
Yes, if you’re building software with Claude Code, Codex, or Cursor, such as an internal tool or a small app. Forge checks each change, asks before anything risky, and ends with a recap of what was checked. It doesn’t replace an expert’s judgment.
Workflows, skills, and rules
A skill is know-how. A workflow is the process. A skill teaches your AI how to do something, and the AI decides when to use it. With a workflow, Forge holds the steps, hands them over one at a time, stops where a person signs off, and keeps a record. Workflows run your skills, so you keep both. See how Forge works with skills →
Yes. For one person doing a short task, that may be all you need. Forge helps when several people must follow the same process, when something needs sign-off, or when someone will ask later what happened. When a skill is enough →
No. Your existing instructions stay as they are. CLAUDE.md, AGENTS.md, .cursorrules, and your skills keep guiding how the AI writes code in a session. Forge works at the layer above: what the work is, what context the AI gets, where it pauses for you, and what gets recorded.
No. Claude Code’s dynamic workflows fan one large job out to many agents. A Forge workflow walks one job through steps and pauses for people to sign off. They solve different problems, and you can use both.
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 keeps one record of what happened.
Working with Forge
No. Forge runs inside Claude Code, Codex, and Cursor, the tools you already use. There is no new app to learn. You start a workflow by saying “forge,” and what you need, at the start of a session or partway through one, and keep working in your tool. The skills and conventions you already have are used inside the steps.
No. You can pull Forge in at any point. Say “forge, fix this” or “forge, review the PR for PROJ-412” forty turns in, and the workflow picks up from where you are: the conversation, the files you opened, and the decisions you made carry into it.
An admin can also switch on session observation. Forge then notices product work happening outside a workflow and offers to track it or pick it up. See how mid-session pickup works →
It is built to take repeated work off you. Forge briefs the AI with your requirements and standards at each step, so you stop re-explaining context in every session. It pauses only where you decided a person should sign off, and only the steps a job needs run.
Yes. Every workflow is a list of steps you control, each with a model tier, an approval setting, and the skills it should use. In the Forge dashboard, an admin switches workflows and steps on or off, sets where a step asks for sign-off, and marks skills as required, for the whole organization or for one team.
To add, reorder, or rewrite steps, describe the change in your AI tool. Forge shows exactly what changes before it saves. On your own account, you are the admin. See how to build your own →
Tools and cost
AI tools: Claude Code, Codex, and Cursor each have a dedicated plugin, and Forge also works with any client that supports MCP, the open Model Context Protocol. The plugins add guardrails: file edits and writes to connected tools wait while a question or an approval is open. Token usage is measured in Claude Code and supported Codex versions, and not in Cursor. Compare the tools →
Jira, Linear, and Notion connect through your AI tool’s own integrations, so results are written to the ticket you already use. Reviewing a pull request, fixing a bug, 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.
Each workflow step is marked fast, balanced, or capable. Routine steps, like fetching a ticket or posting status, go to cheaper, faster models, and hard, code-heavy steps go to the most capable one. It works in both directions: on a top model, routine steps are handed down to a cheaper one, and on a cheaper model, hard steps are sent up. Nobody switches models by hand.
Does it lower cost? That depends on your work and your plan. Token Intelligence shows the estimated spend, so you can check. Model routing works in Claude Code and Codex. In other tools, the workflow uses the model already running your session.
It reads the usage your tools already record — input, output, and cache tokens — with no proxy between you and the model. It converts usage from every measured model and tool into one unit, the “weighted token,” so you can compare them, then prices it at published list rates. It is an estimate: it will not match a flat-rate subscription bill or negotiated pricing.
Getting started
Install the plugin in your AI tool, sign in, and start with one real task. Signing in creates a free account if you don’t have one. Forge offers to switch on the workflow that fits and run it. A team admin can run guided setup to switch on a recommended set. Individual is free forever. Team includes shared workflows, roles, organization-wide analytics, and Token Intelligence for up to 10 members. During beta, Team features are free and each account starts with one seat; request more from your dashboard. After beta, Team is $50/month. Business is coming soon.
Your AI tool handles access to your code. Forge guides the workflow using context that tool sends, which may include code, ticket, document, or message excerpts. Reading a file in your AI tool does not, by itself, send it to Forge. We do not use your data to train AI models.
Forge saves working context so you can resume a run, and team practices and workflow settings so you can reuse them. Some supplied excerpts may be saved with those records. See our Privacy Policy for retention details, security measures, and how to request deletion.