Introduction

Forge is a command-line agent that works inside a repository. You give it a task in one sentence; it explores the codebase, edits files, runs your tests, reads what came back, and keeps going until the task is done or it tells you where it stopped.

The loop

Every task runs through the same cycle. The final stage feeds back into the middle, which is what separates an agent from a single model call.

• Explore - toolchain, layout, and git state, before the first edit.
• Plan - state what is wrong, then make the smallest change that tests it.
• Implement - exact edits, whole files, or unified diffs, confined to the workspace.
• Verify - run the project's own tests and read the exit code.
• Iterate - a failure routes back to explore, up to a hard iteration ceiling.

What it is not

• Not a chat window. It is judged on whether the repository works afterwards.
• Not an editor plugin. It runs in your shell, over ssh, in CI, wherever you are.
• Not a hosted service. There is no account and no Forge server - you supply a model credential and it talks to that endpoint directly.

Install

$ curl -fsSL https://forge.tusharumbarkar.space/install.sh | sh

macOS and Linux. Python 3.11+ is handled for you. See Installation for the other methods.

Where it stands

Forge is version 0.1.0 and alpha. The agent loop, the tool registry, the permission policy, the git tools, and the Docker sandbox all work today. MCP support does not exist yet, and pages that describe unfinished work say so.