Forge — AI Coding Agent Platform
Local-first AI coding agent. VS Code, CLI, Windows, Unity Editor — all powered by your own LLM.
Forge is a local-first AI coding agent built for hands-on developers who want deep control over what the AI does to their code. Every agent action is automatically snapshotted and reversible. The agent runs as a Node.js server on localhost, connecting to Anthropic Claude, OpenAI GPT, or local Ollama models with intelligent routing and automatic failover.
Four interface surfaces share the same Core Engine: VS Code Extension (chat sidebar, inline edit, code actions, diff review, 5 panels), CLI (13 commands, React-based terminal UI, streaming markdown), Windows Desktop (.NET MAUI app with global hotkeys, system tray, overlay window), and Unity Editor (dockable chat panel with 18 scene tools, AI 3D asset generation, deep project intelligence).
The killer feature most coding agents don't have: AI 3D asset generation pipeline. Type "generate a fire-breathing dragon with idle animation" and Forge orchestrates external APIs (Meshy, ElevenLabs, OpenAI, DeepMotion) to produce a 3D model, rig it, animate it, validate quality, and import it into your Unity project with correct ModelImporter settings.
Forge is the same agent system The Klaus Project, LLC uses internally. We built it for our own work, dogfooded it for months, and red-teamed it before releasing. It is the technical receipt behind every AI security audit and threat model we deliver.
What's inside
- Snapshot Engine — automatic pre-action file capture, one-click rollback, session-level undo
- Circuit Breaker — detects stuck agent loops and pauses for human guidance instead of spiraling
- Auto-Verification — runs your build/test command after every file mutation; agent self-corrects on errors
- Multi-LLM routing — Anthropic Claude, OpenAI GPT, or local Ollama with intelligent failover
- Multi-Agent Orchestration (Triad) — 68 tools for coordinating multiple AI agents on the same project
- AI 3D asset pipeline — image→3D→rig→animate→Unity import in one chat command
- VRAM management — automatic Klaus suspend/resume when other models need GPU memory
- MIT licensed — fork it, modify it, run it on your own infrastructure