AI-driven end-to-end full-stack development workbench
From requirement clarification to deployment verification — all through conversation
OpenForge is an AI-powered workbench that turns conversation into production code. Describe what you want to build, and OpenForge handles the entire pipeline — requirements analysis, architecture design, implementation, testing, deployment, and verification.
Built with Conduit (RealWorld monorepo) as the experimental field.
You describe → PM Agent → Pipeline Engine → Deployed Code
"I want a clarifies, builds, tests, running on
blog app" decomposes, reviews, your infra
designs deploys
┌──────────────────────────────────────────────────────────┐
│ C-Layer · Workbench │
│ React + Dockview + WebSocket │
├──────────────────────────────────────────────────────────┤
│ A-Layer · Pipeline Engine │
│ Go · State Machine · Gate · Deploy │
├──────────────────────────────────────────────────────────┤
│ B-Layer · Agent Swarm │
│ Go + Node.js · CSP Channels · Multi-Agent │
├──────────────────────────────────────────────────────────┤
│ Infrastructure │
│ PostgreSQL · Redis Streams · Docker │
└──────────────────────────────────────────────────────────┘
| Module | Capability |
|---|---|
| Conversational PM | Natural language → structured requirements → task decomposition |
| Pipeline Engine | 9-state transition machine with approval gates and backtracking |
| Agent Swarm | Multi-agent coordinator with Spawn / Delegate / Broadcast patterns |
| Tool Registry | 7 core tools — File I/O, Code Search, Shell Exec, DB Query, Git, Web Fetch |
| Pro Mode | Full IDE: Chat + Diff + File Tree + Terminal + Topology in Dockview panels |
| LLM Router | Multi-provider: Anthropic / OpenAI / DeepSeek, with token metering |
| Sandbox | Docker container isolation with LRU warm pool and 5-layer defense |
| Compliance | WORM audit logs, data lifecycle, monthly partitioning |
- Go 1.22+ · Node.js 20+ · PostgreSQL 15+ · Docker (optional)
# 1. Clone
git clone https://github.com/Yorha9e/openforge.git && cd openforge
# 2. Configure
cp config/profiles/minimal.yaml.example config/profiles/minimal.yaml
cp config/profiles/docker-dev.yaml.example config/profiles/docker-dev.yaml
# 3. Environment
export DB_PASSWORD=your_db_password
export JWT_SECRET=your_jwt_secret_at_least_32_chars
export ANTHROPIC_AUTH_TOKEN=your_anthropic_api_key
# 4. Start database
docker compose up postgres -d
# 5. Run migrations
go run cmd/openforge/main.go migrate up
# 6. Start server
go run cmd/server/main.go --addr :8030
# 7. Start frontend
cd frontend && npm install && npm run devOpen http://localhost:5173 — start building through conversation.
openforge/
├── cmd/
│ ├── server/ REST + WebSocket API server
│ └── openforge/ CLI entry point
├── proto/ Protocol Buffers (6 services)
│
├── internal/
│ ├── agent/ AI agent: coordinator, query engine, tools
│ ├── pipeline/ Pipeline: state machine, gate, deploy
│ ├── auth/ Authentication: JWT, RBAC, OIDC
│ ├── llm/ LLM: provider router, translator, token meter
│ ├── observability/ Monitoring: Prometheus, circuit breaker
│ └── shared/ Shared kernel, profiles, feature flags
│
├── frontend/ React 18 + TypeScript + Vite
│ └── src/features/ chat / code-review / admin / monitoring
│
├── nodejs-io/ Node.js I/O service (gRPC)
├── migrations/ Database migrations (PostgreSQL)
├── config/profiles/ Capability profiles (YAML)
├── deployments/ Docker, Nginx, TLS configs
└── scripts/ Diagnostic & utility scripts
Three tiers, one codebase:
| Profile | Target | Key Traits |
|---|---|---|
minimal |
<10 people, single machine | SQLite / local FS / in-memory |
standard |
50-200 people, single-AZ K8s | PG / Redis / MinIO / Vault |
enterprise |
Regulated, multi-AZ | HA + DR + WORM + Feishu + OIDC |
# Switch profile
export OPENFORGE_PROFILE=standard
go run cmd/server/main.go| Layer | Technology |
|---|---|
| Backend | Go · gRPC · gorilla/websocket · bcrypt · golang-jwt |
| Frontend | React 18 · TypeScript · Vite · Dockview · Monaco Editor |
| Database | PostgreSQL 15+ · UUID v7 · WORM partitioning · 22 tables |
| Infrastructure | Docker Compose · Nginx · TLS · Redis Streams (Phase 5+) |
| Protocol | gRPC · WebSocket (14 upstream + 14 downstream events) |
| AI/LLM | Anthropic Claude · OpenAI · DeepSeek · Multi-provider router |
# Run all tests
go test ./...
# Frontend tests
cd frontend && npm test
# Build
go build -o bin/server cmd/server/main.go
cd frontend && npm run build
# Lint
golangci-lint run
cd frontend && npm run lintNever commit:
.envfiles · API keys · Private keys · Certificates- Configuration files with hardcoded credentials
- Password hashes (use
htpasswd -nbBC 10 "" <password>)
Use environment variables for all secrets. See .env → .env.example pattern.
为响应 DESIGN §19.4 + 8 子代理审计识别的 31 stub（25 P0 + 35 P1 + 30 P2），我们执行了 6 路径补完计划：
| Plan | Tasks | Branch | PR | Status |
|---|---|---|---|---|
| Path A: Phase 4 数据闭环 | 11 | feat/path-A-data-closure |
#3 | ✅ 完成 |
| Path B: 安全 + UX 闭环 | 11 | feat/path-B-security-ux-closure |
#1 | ✅ 完成 |
| Path C: 架构兑现 | 14 | feat/path-C-architecture-honesty |
#5 | ✅ 完成 |
| Path D: Enterprise 落地 | 11 | feat/path-D-enterprise-landing |
#6 | ✅ 完成 |
| X3: DB 24 项优化 | 9 | feat/X3-db-24-optimization |
#2 | ✅ 完成 |
| X4: 契约 + Chaos + Bench | 5 | feat/X4-proto-fault-bench |
#4 | ✅ 完成 |
关键产物：
- 22 个 gRPC RPC 真实启用（Coordinator/Gate/LLM/Tools/Learning/Terminal）
- Token 数据流端到端真实落库（Node.js → gRPC → PG
token_usage） - WORM 审计链 hourly 扫描 + 链断裂告警
- 13 metric 全部上报 Prometheus
- 18 条告警规则 + alertmanager feishu webhook
- 10 个 Playwright E2E 用例
- RLS 行级安全 + UUID v7 + 列级加密
- 4 类 Benchmark 监控 + 5 类 Chaos 故障注入
- L1-L4 自学习 4 层闭环
- 6 个 worktree 隔离 + 6 份 subagent dispatch prompt 模板
计划文档（docs/superpowers/plans/）：
INDEX.md— 6 计划依赖图2026-06-06-DESIGN-stub-todo.md— 总体 stub 清单2026-06-06-path-{A,B,C,D}-*.md— 4 路径 plan2026-06-06-X{3,4}-*.md— 2 跨路径 plan2026-06-06-{X3,path-B,path-C,path-D,X4}-dispatch-prompts.md— 5 份 subagent 派发模板
审计报告：
- 整体完成度 ~51% → ~85% （代码层）+ 7 个 P0 必修 stub 全部关闭
- 详细 see
2026-06-06-DESIGN-stub-todo.md头部 5 段 supersession
To be determined