Multi-agent inside a single tool works fine. Claude Code runs 20 subagents and they stay aligned because the tool manages that coordination internally.
The problem is multi-tool. Claude Code, Codex CLI, and Gemini CLI on the same repo with no shared state, no file locking, no knowledge capture. Each tool starts from scratch. Decisions evaporate between sessions. Two tools edit the same file and neither knows the other exists.
Forge solves this. One product, three depths.
L1 — The Safety Net: Forge PluginZero-dependency governance for Claude Code. Install in 30 seconds. Health scoring, gap analysis, quality gates, and specialized agents. claude plugin add nxtg-ai/forge-plugin21 commands; 22 agents; 29 skills; 6 hooks; 8 MCP tools. |
L2 — The Delivery Engine: Forge OrchestratorMulti-tool coordination in a single Rust binary. File locking, knowledge capture, task planning, drift detection. Coordinates Claude Code, Codex CLI, and Gemini CLI. curl -fsSL https://forge.nxtg.ai/install.sh | sh
forge init4MB binary; 292 tests; 10 MCP tools. |
L3 — Mission Control: Forge DashboardFull visual platform for multi-tool AI coordination. Real-time governance HUD, agent activity feed, and the Infinity Terminal: sessions that survive browser close, network drops, and device switches. git clone https://github.com/nxtg-ai/forge-ui
npm install && npm run dev58 components; 4,146 tests; 87% coverage. |
forge-orchestrator (Rust, 4MB)
Policy core. File locking. Knowledge flywheel.
Governance. MCP server. Multi-tool adapters.
Policy enforced here. Nowhere else.
| |
forge-plugin forge-ui
(L1: Safety Net) (L3: Mission Control)
Claude Code React dashboard
21 commands Infinity Terminal
22 agents Governance HUD
29 skills Agent feed
Communication: .forge/ filesystem + MCP stdio. No daemon. No database. State is files.
Every developer running multiple AI tools on the same repo hits the same walls.
File locking. Two tools editing the same file at the same time. No locks. No coordination. Forge acquires exclusive locks per file; queued tools get notified.
Knowledge capture. Decisions, patterns, learnings captured automatically. Auto-classified and searchable. A Claude Code session's decisions are available to Codex CLI the next day.
Drift detection. Your spec says "build auth" but the tool is refactoring CSS. Forge catches divergence early.
Governance. Health scores from A through F, computed across 8 quality dimensions. Continuously validated, not just at PR time.
These are the same failure modes that break billion-dollar enterprise programs. The founder spent 23 years solving them with human teams. Forge encodes those patterns for AI tools.
# L1: Governance in 30 seconds
claude plugin add nxtg-ai/forge-plugin
# L2: Multi-tool coordination
curl -fsSL https://forge.nxtg.ai/install.sh | sh
forge init
# L3: Visual dashboard
git clone https://github.com/nxtg-ai/forge-ui
npm install && npm run devEach depth builds on the last. Nothing forces you to go deeper. Adoption follows the pain.
| Repo | Level | Language | Tests |
|---|---|---|---|
| forge-plugin | L1: The Safety Net | Markdown + Shell | Structure validation |
| forge-orchestrator | L2: The Delivery Engine | Rust | 292 |
| forge-ui | L3: Mission Control | TypeScript + React | 4,146 |
forge.nxtg.ai | nxtg.ai | Built by @AxW
"The delivery engine your AI tools are missing."