Skip to main content
Product•4 min read

Forge is Live: Your AI Tools Deserve a Tech Lead

February 9, 2026 by Asif Waliuddin

ForgeLaunchAI OrchestrationOpen SourceClaude CodeCodex CLIGemini CLIMCPRust
Forge — Your AI tools deserve a tech lead

Your AI tools are powerful. Together, they're chaos.

You have Claude Code. You have Codex CLI. You have Gemini CLI. Each one can read your codebase, reason about architecture, and ship production code autonomously.

But run two of them on the same repo? You get merge conflicts. Lost context. Silent architectural drift. One agent refactors your auth while the other builds features against the old pattern. Nobody remembers what was decided yesterday.

This is not a tooling problem. It is a coordination problem. And today, we are shipping the solution.

Introducing Forge

Forge is a 3 MB Rust binary that turns your AI coding tools into a coordinated development team. Think of it as a senior tech lead that never sleeps — it plans work, assigns it to the right AI, prevents file conflicts, captures decisions, and catches drift before it compounds.

curl -sSL https://raw.githubusercontent.com/nxtg-ai/forge-orchestrator/main/install.sh | sh

That's it. One command. Zero runtime dependencies. Works on Linux, macOS, and Windows.

What Forge Actually Does

1. Plans work from your spec

Write a SPEC.md describing what you want to build. Forge decomposes it into tasks with dependencies, file assignments, and agent recommendations — powered by GPT-4.1.

forge plan --generate

2. Assigns the right AI to each task

Not all AI tools are equal at every task. Forge's detection engine identifies which tools are installed and available, then assigns tasks based on complexity and tool strengths.

3. Prevents file conflicts automatically

When an agent claims a task, Forge locks the associated files. No other agent can touch them until the task is complete. This is the single most important feature — it eliminates the merge conflict chaos that plagues multi-agent workflows.

4. Captures knowledge that persists

Every decision, pattern, and learning gets stored in a searchable knowledge base. When Agent B starts working tomorrow, it has access to everything Agent A discovered today. The knowledge flywheel gets more valuable with every session.

5. Detects architectural drift

Forge compares completed work against your original spec using LLM-powered analysis. If an agent has wandered off-plan, you know immediately — not three days later during code review.

The 60-Second Experience

# Install
curl -sSL https://raw.githubusercontent.com/nxtg-ai/forge-orchestrator/main/install.sh | sh
 
# Initialize in any git repo
cd your-project
forge init
 
# Write your spec, generate a plan
forge plan --generate
 
# Ship it — all agents work in parallel
forge start

forge start is the headline feature. It launches all available AI tools in parallel, auto-claims tasks, handles retries on transient errors, and completes when every task is done. Add --ceo for fully autonomous mode — it loops until the entire plan is shipped.

Built on MCP

Forge includes a built-in Model Context Protocol server with 9 tools. Add it to your Claude Code config:

{
  "mcpServers": {
    "forge": {
      "type": "stdio",
      "command": "forge",
      "args": ["mcp", "--project", "."]
    }
  }
}

Now Claude can call forge_get_tasks(), forge_claim_task(), forge_check_drift() natively. The orchestration state is live and queryable from inside your AI tools.

What We Tested

Before shipping, we ran three UAT journeys:

  • Journey 1 — Orchestrator solo: End-to-end autonomous task execution across Claude, Codex, and Gemini. Found and fixed 10 friction points including installation path, error retry patterns, and MCP configuration.
  • Journey 2 — Plugin integration: Validated 20 slash commands, 22 agent specs, and 29 skills work standalone without a running server. Fixed the initialization flow to use native Claude Code tools.
  • Journey 3 — UI dashboard: Tested all 6 pages of the React dashboard with Playwright. Found and fixed a self-DoS bug (17 polling endpoints exceeding rate limits). Zero console errors after fix.

Why Open Source

Multi-AI orchestration is infrastructure. It should not be proprietary. Every developer running more than one AI tool hits the same coordination problems. The solution should be free, auditable, and extensible.

Forge is MIT-licensed. The entire codebase — Rust binary, MCP server, test suite — is open.

Try It Now

curl -sSL https://raw.githubusercontent.com/nxtg-ai/forge-orchestrator/main/install.sh | sh

Your AI tools are already capable. Give them a tech lead.