Open-source local AI coding tools · Your files stay on your machine · Explore on GitHub ↗
Implementation

A small bridge between
your AI and your code.

SFI Agent runs locally. Remote connectivity is an optional network path—not a hosted execution service.

Components

Request flow

  1. The client connects to the configured local or remote MCP endpoint.
  2. The MCP server negotiates protocol capabilities and lists available tools.
  3. A user request may result in one or more tool calls.
  4. The local host validates the tool and its arguments against the configured workspace and permissions.
  5. The tool returns a result or a structured error to the client.
  6. The developer reviews code changes, test results, and Git diffs.

Deployment model

The public SFI Agent website distributes documentation and installer scripts. The actual coding tools should run on the user's own workstation. A Cloudflare Tunnel or ngrok can provide remote reachability, but should forward only the necessary endpoint. Do not confuse the marketing site's health status with the status of a user's local workstation.

Implementation and testing checklist