A web platform for creating AI-toolchain projects. Describe your project — agent-planforge and scaffoldkit do the rest.

Live: project-forge.opentriologue.ai

project-forge resolves generated planforge artifacts via planforge-index.json when available and falls back to legacy root paths for older planforge installations.

1. Describe — Fill in a form (or use the ✨ AI magic fill)
2. Review — Browse generated tasks, architecture overview, and file tree
3. Confirm — Create a GitHub repo with the scaffold pushed
4. Build — Clone and hand off to your agent

AI is optional. Without it, project-forge uses deterministic intake mapping plus agent-planforge heuristics. If a local or hosted AI provider is configured, project-forge also uses it server-side to enrich intake and review scaffold fit after scaffoldkit runs.

Also available as a REST API for agents — see the API section below.

project-forge talks to a single dependency: the agent-planforge HTTP service, which runs both the planforge CLI and scaffoldkit in its own container and returns the scaffolded project as a tarball over HTTP (per ADR-0002). Docker Compose wires it in automatically; nothing to install on the host.

git clone https://github.com/LanNguyenSi/project-forge.git
cd project-forge
cp .env.example .env
# Fill in required values (see below)
make deploy

Publishing a repo uses each user's own GitHub Personal Access Token (PAT), added in the dashboard, not a platform-wide token. There is no server-level GITHUB_TOKEN or GITHUB_OWNER env var. For OAuth sign-in, set the optional GITHUB_ID / GITHUB_SECRET below.

The root docker-compose.yml ships two services:

• app — the project-forge Next.js runtime.
• planforge: the agent-planforge HTTP service (per ADR-0002). Built from the sibling agent-planforge checkout via context: ../agent-planforge (see docker-compose.yml). Runs both the planforge CLI and scaffoldkit in-container. Internal-only, no Traefik labels, no published ports. app reaches it via the shared traefik docker network at http://planforge:8223.

Both services need PLANFORGE_SERVICE_TOKEN in .env. Compose propagates it; mismatched values cause app to get 401s from planforge.

Token rotation (manual, v1):

cd /root/git/project-forge
NEW_TOKEN=$(openssl rand -hex 32)
sed -i "s/^PLANFORGE_SERVICE_TOKEN=.*/PLANFORGE_SERVICE_TOKEN=$NEW_TOKEN/" .env
docker compose up -d --build  # rebuild both so the new token is live simultaneously

A token store + in-place rotation is a follow-up (see ADR-0002).

All endpoints require an API token generated in the dashboard, passed via the X-API-Key header.

Rate limit: 10 project publishes per user per day. Only publish and the one-shot POST /api/v1/projects count against it; generate, preview, and the project list/delete endpoints are unmetered. The quota is counted per user, across all of that user's tokens.

List all projects for the authenticated user.

curl https://project-forge.opentriologue.ai/api/v1/projects \
  -H "X-API-Key: pf_your_token_here"

Soft-delete a project by its id (the id field returned by GET /api/v1/projects), passed as a query parameter.

curl -X DELETE "https://project-forge.opentriologue.ai/api/v1/projects?id=clx..." \
  -H "X-API-Key: pf_your_token_here"

Generate a project scaffold without publishing it. Returns a sessionId (a UUID) used by preview and publish. A session expires one hour after it is generated.

curl -X POST https://project-forge.opentriologue.ai/api/v1/generate \
  -H "X-API-Key: pf_your_token_here" \
  -H "Content-Type: application/json" \
  -d '{
    "projectName": "my-cli-tool",
    "summary": "A CLI that syncs agent memory via Git",
    "features": ["push memory files", "pull and merge", "conflict resolution"],
    "constraints": ["TypeScript only", "no external databases"],
    "targetUsers": ["developers", "AI agents"]
  }'

Response:

{
  "ok": true,
  "sessionId": "f47ac10b-58cc-4372-a567-0e02b2c3d479",
  "preview": { "...": "file tree, tasks, architecture overview" }
}

Fetch the generated preview (file tree, tasks, architecture) for a given sessionId (the value returned by generate).

curl "https://project-forge.opentriologue.ai/api/v1/preview?sessionId=f47ac10b-58cc-4372-a567-0e02b2c3d479" \
  -H "X-API-Key: pf_your_token_here"

Finalize a previewed project and create the GitHub repository. Requires a GitHub PAT configured in your dashboard, and counts against your daily publish quota.

curl -X POST https://project-forge.opentriologue.ai/api/v1/publish \
  -H "X-API-Key: pf_your_token_here" \
  -H "Content-Type: application/json" \
  -d '{ "sessionId": "f47ac10b-58cc-4372-a567-0e02b2c3d479" }'

Response:

{
  "ok": true,
  "result": {
    "repoUrl": "https://github.com/your-username/my-cli-tool",
    "cloneUrl": "https://github.com/your-username/my-cli-tool.git",
    "projectName": "my-cli-tool"
  }
}

One-shot create: generate, scaffold, create the GitHub repository, and push in a single call, skipping the separate preview step. Requires a GitHub PAT configured in your dashboard, and counts against your daily publish quota. The request body is the same shape as generate.

curl -X POST https://project-forge.opentriologue.ai/api/v1/projects \
  -H "X-API-Key: pf_your_token_here" \
  -H "Content-Type: application/json" \
  -d '{
    "projectName": "my-cli-tool",
    "summary": "A CLI that syncs agent memory via Git",
    "features": ["push memory files", "pull and merge"]
  }'

Returns the same { ok, result: { repoUrl, cloneUrl, projectName } } shape as publish.

Full API documentation: project-forge.opentriologue.ai/docs

• Frontend: Next.js 15 + TypeScript + Tailwind CSS
• Auth: next-auth (email/password + GitHub OAuth)
• Database: SQLite via Prisma (API tokens, users)
• Planning + Scaffolding: agent-planforge HTTP service (bundles scaffoldkit in its own container; project-forge is Node-only)
• AI: Local OpenAI-compatible endpoint, Groq (llama-3.3-70b-versatile), or OpenAI (gpt-4o-mini)
• Deploy: Docker + Traefik

npm install
npm run dev

MIT