Product

Everything in the editor.
Nothing in your way.

Goatfied is a full IDE, an agent runtime and a model gateway in one product. Here is every part of it.

Goat-Tab Cmd+K Agents Codebase MCP Teams

Goat-Tab

It predicts the edit,
not just the token.

Goat-Tab is a purpose-trained edit-prediction model served from warm GPUs inside our own VPC. It watches the shape of your recent changes and proposes the whole next edit — across lines, across files, in one keystroke.

Multi-line and multi-file diffs from one Tab press
Jumps to the next place the edit needs to land
Learns your project conventions from the index
Falls back to local heuristics when you're offline

pricing.tsgoat-tab

  1  export function applyDiscount(cents: number) {
  2    const rate = cents > 20_000 ? 0.2 : 0.1;
  3    return Math.round(cents * (1 - rate));
  4  }

Goat-Tab proposes the discount-rate branch and the rounded return value together; one Tab press accepts both lines.

users.pycmd+k

make this paginated and add a total count

  1  def list_users(db, page=1, size=50):
  2      total = db.query(User).count()
  3      return db.query(User).offset((page - 1) * size).limit(size)

A plain-English instruction becomes a reviewable inline diff — accept, reject or refine before anything touches the file.

Cmd+K & Chat

Describe the change.
Review the diff.

Select code, press Cmd+K, say what you want. Goatfied streams an inline diff you approve hunk by hunk. Cmd+L opens the same model in a chat pane with your whole codebase as context.

Inline diffs, never silent overwrites
Frontier model routing on hard changes
Works in any file — code, config, SQL, Markdown
@-mention files, symbols, docs and MCP tools

Agents

Hand off a ticket. Get a pull request.

Background agents clone the repo into a sandbox, reproduce the issue, patch it, run your tests, and open a PR. You review code, not chat logs.

Real pull requests

Agents run in an isolated sandbox with scoped tokens, then push a branch and open a PR you review like any other.

Replayable traces

Every tool call, file read, diff and command is recorded. Open the trace and see precisely why the agent did what it did.

Scoped credentials

Agents get a short-lived token narrowed to the repos you name. No org-wide PAT, no standing access.

Drive from CI

Trigger the same agents from GitHub Actions, GitLab CI or your shell with the Goatfied CLI.

Context

It has actually read your codebase.

Retrieval quality is the whole game. Goatfied indexes your repo continuously so the model answers from your code instead of pattern-matching the internet.

Codebase indexing

Every repo is chunked, embedded and kept warm in pgvector. Ask about code you've never opened and get an answer grounded in the actual file, not a guess.

Tree-sitter aware

Syntax-aware chunking across 200+ grammars means retrieved context lands on whole functions and types, not arbitrary 40-line windows.

Incremental re-index

Index updates follow your file watcher and your branch. Switch branches and context follows you — no full re-scan, no stale symbols.

Rules and memories

Project rules in .goatfied/rules travel with the repo, so every engineer and every agent works from the same conventions.

Extend

Your tools. Your models. Your metal.

Goatfied is deliberately not a walled garden. Bring your extensions, your MCP servers, your model keys and, if you want, your own GPUs.

MCP-native

Wire any Model Context Protocol server into chat in one click. Your internal tools become agent tools without writing a plugin.

Open VSX marketplace

A self-hosted Open VSX mirror means your existing VS Code extensions, themes and keymaps come with you.

Model routing

Route fast edits to a small warm model and hard reasoning to a frontier model automatically — or pin a model per workspace.

Bring your own GPU

Self-host the inference plane on Kubernetes for air-gapped teams, or point Goatfied at your own keys.

Teams & Enterprise

Controls your security team will actually accept.

Everything an organisation needs to roll Goatfied out without a six-month review: enforced privacy, directory sync, spend controls and your own infrastructure.

Privacy mode

No training on your code, no retention of prompts or completions. Enforced org-wide, not per-seat.

SSO and SCIM

SAML or OIDC single sign-on with directory-driven provisioning and de-provisioning.

Usage analytics

See acceptance rates, spend and model mix per team. Set hard budget caps before the invoice, not after.

Your VPC, your keys

Run the inference plane inside your own AWS account with customer-managed KMS keys on Enterprise.

Surfaces

One account. Everywhere you work.

Desktop

Native app for macOS, Windows and Linux. Full local filesystem, terminal, debugger and extensions.

Download

Web

The same workspace in a browser tab. Nothing to install — useful for reviews, quick fixes and locked-down machines.

Open in browser

CLI

Script agents, run evals and wire Goatfied into CI. Same auth, same models, same traces.

Read the docs

Try the whole thing. Free.

Every feature above is available on the free plan with limits. No card required.

Download Goatfied See pricing

Everything in the editor.Nothing in your way.

It predicts the edit,not just the token.

Describe the change.Review the diff.