Convolution Labs
Build AI agents and the data pipelines that feed them, on a visual canvas that guides you toward the secure, production-ready version by default.
You wire nodes together — agent, tool, condition, retrieval, input, output — and the graph compiles to a real execution pipeline: topologically sorted, run on a queue-backed worker, with every node's progress streamed live so you can watch a run happen and see exactly where it failed.
Start here
| Quickstart | Build, test and publish your first agent |
| Core concepts | Agents, runs, credits, and how they fit together |
Building agents
| The canvas | The visual editor, end to end |
| Node reference | Every node type and what it does |
| Agent rules | Typed rules that compile into your prompts |
| Tools | Built-in tools and the Python sandbox |
| Retrieval (RAG) | Ground agents on your own documents |
Running in production
| Security scanning | Find prompt injection, SSRF and excessive agency before shipping |
| Data pipelines | Scheduled syncs that feed your agents |
| Run an agent via API | Publish an agent and call it from your code |
| Billing | How credits are priced and consumed |
| Teams | Roles, invitations and shared workspaces |
What you can build
- Agents — multi-step LLM workflows with tool calling, branching, retrieval and custom Python, starting from built-in patterns (ReAct, chain-of-thought, self-reflection) rather than a blank canvas.
- Data pipelines — scheduled Source → Destination syncs with incremental loads and SQL transforms, so your agents have data to reach for.
- Security scans — static analysis of an agent's graph plus dynamic probes that attack it through its own execution path, scored and mapped to the OWASP LLM Top 10 and NIST AI RMF.
:::note Bring your own keys Agent runs use your LLM provider keys (OpenAI, Anthropic, Gemini), stored encrypted and decrypted only inside the worker. Add them under Dashboard → Settings → LLM Keys before your first run. :::