CrewBuilder

Visual platform for building, running, and observing CrewAI agents. No YAML. No direct CrewAI dependency in the UI.

Stack

| Layer | Technology | |-------|-----------| | Frontend | Vite 5 + React 18 + TypeScript 5 + Tailwind CSS 3 + TanStack Query v5 | | Backend | FastAPI + SQLAlchemy + SQLite | | Agent runtime | CrewAI ≥1.0 (engine/) | | Deploy | Docker (single container) |

Features

• AI Wizard — generates complete projects (agents + tasks) from a natural language description
• Post-execution approval flow — first task ("Preliminary Proposal") pauses execution and waits for user sign-off before continuing
• Edit & retry — user can edit a proposal output before approving it
• Task-level checkpoints — pause and resume mid-run at any task boundary
• Result downloads — per-step .md download and a full .zip of all steps
• Real-time streaming — execution events transmitted via SSE
• Observability — per-agent steps with token usage, status, and expandable output
• Knowledge bases — PDFs, text, URLs indexed with BM25 + native CrewAI RAG
• Memory — STM / LTM / Entity memory scoped per project via ChromaDB
• Guardrails — per-task output validators
• Agent training — crew.train() via dedicated endpoint
• Triggers — schedule or webhook-triggered runs
• Agent templates — reusable agent definitions
• Public API tokens — per-project API tokens for external access
• Langfuse tracing — optional, zero-config via env vars
• Preflight checks — structural validation + tool instantiation before a run

Local development

Requirements

• Node.js 20+
• Python 3.11+
• Docker (production only)

Quick start

# Install dependencies
cd apps/web && npm install
cd ../api && pip install -r requirements.txt
pip install -e ../../engine/

# Start backend + frontend in parallel
bash scripts/dev.sh

• Frontend: http://localhost:5173
• API: http://localhost:8000
• API docs: http://localhost:8000/docs

API keys

Keys are stored in ~/.crewbuilder/.env and loaded automatically at startup:

ANTHROPIC_API_KEY=sk-ant-...
OPENAI_API_KEY=sk-...
SERPER_API_KEY=...

They can also be configured from UI → Settings → API Keys.

Backend only

cd apps/api
uvicorn app.main:app --reload --port 8000

Frontend only

cd apps/web
npm run dev

Docker

docker compose -f docker/docker-compose.yml up -d

App available at http://localhost:8000

Build

bash scripts/build.sh

Project structure

├── apps/
│   ├── web/        # React frontend (Vite + Tailwind)
│   └── api/        # FastAPI backend
├── engine/         # crewbuilder-engine (Python package)
├── data/           # SQLite DB (gitignored)
├── scripts/        # dev.sh, build.sh, start.sh
└── docker/         # Dockerfile + docker-compose.yml

API reference

Projects

GET    /api/projects
POST   /api/projects
GET    /api/projects/:id
DELETE /api/projects/:id
GET    /api/projects/:id/agents
POST   /api/projects/:id/agents
DELETE /api/projects/:id/agents/:agent_id
GET    /api/projects/:id/tasks
GET    /api/projects/:id/runs
GET    /api/projects/:id/preflight
GET    /api/projects/:id/export/python
POST   /api/projects/:id/train
POST   /api/projects/:id/api-token
DELETE /api/projects/:id/api-token
GET    /api/projects/:id/evaluations/summary
GET    /api/projects/:id/feedback
POST   /api/projects/:id/feedback
POST   /api/projects/:id/feedback/:feedback_id/approve

Runs

POST   /api/runs
GET    /api/runs/recent
GET    /api/runs/:id
GET    /api/runs/:id/events
GET    /api/runs/:id/steps
GET    /api/runs/:id/stream          # SSE
GET    /api/runs/:id/checkpoint
POST   /api/runs/:id/approve         # Approve a task checkpoint
POST   /api/runs/:id/approve-output  # Approve proposal and continue
POST   /api/runs/:id/resume
POST   /api/runs/:id/cancel
POST   /api/runs/:id/replay

Tools, Knowledge, Memory

GET    /api/tools
GET    /api/tools/connections
GET    /api/tools/custom
POST   /api/tools/config
GET    /api/knowledge
POST   /api/knowledge/upload
POST   /api/knowledge/text
POST   /api/knowledge/url
POST   /api/knowledge/search
GET    /api/knowledge/:id/preview
POST   /api/knowledge/:id/reindex
PATCH  /api/knowledge/:id
DELETE /api/knowledge/:id
GET    /api/memory
DELETE /api/memory

AI, Skills, Triggers, Settings

POST   /api/ai/generate-project
POST   /api/ai/generate-agent
POST   /api/ai/refine-prompt
POST   /api/ai/review-tasks
POST   /api/ai/fix-tasks
GET    /api/skills
POST   /api/skills
PATCH  /api/skills/:id
DELETE /api/skills/:id
GET    /api/triggers
PATCH  /api/triggers/:id
DELETE /api/triggers/:id
GET    /api/agent-templates
POST   /api/agent-templates
DELETE /api/agent-templates/:id
GET    /api/settings/llm-keys
GET    /api/settings/local-path
GET    /api/settings/aider
GET    /api/settings/runtime-guardrails
GET    /api/settings/tool-defaults
GET    /api/events
GET    /api/health

Implementation status

| Feature | Status | |---------|--------| | Domain models + schemas | ✅ | | SQLite DB + Alembic migrations | ✅ | | Full REST API | ✅ | | Full UI (Observatorio design) | ✅ | | Engine + compiler | ✅ | | CrewAI integration | ✅ | | Async execution + SSE streaming | ✅ | | AI Wizard (generate-project) | ✅ | | Post-execution approval flow | ✅ | | Task checkpoints + resume | ✅ | | ZIP download of results | ✅ | | Langfuse tracing (optional) | ✅ | | Agent training | ✅ | | Triggers | ✅ | | Agent templates | ✅ | | Public API tokens | ✅ | | Run diff (compare outputs across runs) | ⏳ spec ready | | Docker (optimized single-container build) | ⏳ pending |