Skip to main content
Private preview. fremforge is in private preview — invited customers only. Content is still subject to change. Request access →
AI in your IDE

AI in your IDE

fremforge stays out of your IDE. AI coding assistants — Claude Code, Cursor, Continue, JetBrains AI Assistant, GitHub Copilot Workspace, Codeium, and others — talk to a Git host the same way any developer tool does: HTTPS Git for clones + pushes, plus the host’s REST API for everything else. fremforge ships the standard pieces — PATs, REST API, OpenAPI spec — so any AI tool that already knows how to talk to GitHub or GitLab works with fremforge with three lines of config.

What this page is, and isn’t

This page is the per-tool quick-start. For each tool you get: which PAT scopes to request, where to paste the fremforge base URL, and a one-liner test that confirms the tool can read your repos. It is NOT a recommendation of which tool to use — fremforge is deliberately tool-agnostic.

For the server-side AI features (AI PR review, Renovate explanations, customer-callable /ai/complete for your own workflows), see the AI integrations admin page inside your org. That’s a separate surface — different audience, different sovereignty story. Quick reframe:

SurfaceAudiencefremforge sub-processes your code?
This page — IDE AI assistantIndividual developersNo — your tool talks to your AI vendor directly with your key
AI integrations admin — PR review, /ai/completeOrg adminsNo — BYOK, your tool, your contract, your vendor

Either path, fremforge never sub-processes customer code through a US AI vendor on the default — see the trust page for the full data-flow diagram.

How fremforge looks to an AI tool

Every AI dev tool that talks to “GitHub” really wants three things:

  1. HTTPS Git for git clone / git push. fremforge ships this at https://frem.sh/<org>/<repo>.git.
  2. A PAT with scopes for the actions it wants to take. fremforge PATs use the prefix ffp_… and the same scope vocabulary (repo:read, repo:write, issues:read, etc.) as GitHub fine-grained PATs.
  3. A REST API with an OpenAPI spec. fremforge’s API is at https://frem.sh/_app/api/v1/* and the spec lives at https://docs.frem.sh/api/openapi.yaml.

If the tool has a “self-hosted GitHub Enterprise” option, point it at fremforge. Most tools do. The per-tool pages below walk through the exact UI clicks.

Minting a PAT for your AI tool

  1. Open https://frem.sh/-/settings/tokens (replace frem.sh with your fremforge host if self-hosted).
  2. Click New PAT.
  3. Name it after the tool (e.g. claude-code, cursor) so revocation later is straightforward.
  4. Pick scopes from the per-tool tables below — start narrow, add scopes only when the tool errors out.
  5. Pick the shortest TTL that fits your workflow (90 days is the platform max).
  6. Copy the ffp_… token. You see it once.

A PAT minted this way is what every per-tool guide below assumes. The scope tables are exact — the tool errors out clearly if a scope is missing.

Per-tool guides

If your tool isn’t listed but supports GitHub Enterprise Server, the pattern is: base URL = https://frem.sh/api/v1, token = your ffp_… PAT, and you’re done. Open an issue at frem.sh/fremforge/docs/issues and we’ll add your tool to this list.