Drag, connect, and run AI agent workflows from a visual canvas. Works with any major LLM — OpenAI, Anthropic, Gemini, and more. Parallel execution, live streaming, and a public API — no code required.
2,400+
workflows built
18M+
tool calls executed
99.9%
uptime
run #a3f9 · customer-support
The idea
Every agent is just LLM calls, tool calls, and routing logic. Pipecat puts that graph on a canvas where you can see, edit, and debug every edge in real time.
Nodes with no shared dependencies run concurrently. A three-branch graph finishes in the time of its slowest branch — not the sum of all three.
Enable a workflow, get a POST /flows/{slug}/invoke endpoint with an API key. Sync, async, or SSE stream. Call it from anywhere in one line of curl.
Capabilities
Drag and drop nodes onto an infinite canvas. Connect them into directed acyclic graphs that execute in topological order.
WebSocket events stream live node status directly to the canvas as each step completes.
POST https://api.your-tool.com/run
{
"query": "{input}"
"limit": 10
}
// result returned to the agent
Register any HTTP endpoint as a tool. The agent calls it automatically mid-run — no extra code.
asyncio.gather — 3x faster
Nodes without dependencies run concurrently, cutting total latency to the slowest branch — not the sum.
$ curl -X POST
https://pipecat.dev/flows/support/invoke
-H "X-Api-Key: sk-…"
-d '{"input": "Where is my order?"}'
// → { result: "Your order ships…" }
Deploy any workflow and get a public REST endpoint with API-key auth. Invoke sync, async, or via SSE stream from any system.
Pricing
Start free. Scale when you're ready.
Hobby
For side projects and exploration.
Pro
For teams shipping production AI.
Enterprise
Unlimited scale, dedicated infrastructure.
No infrastructure to manage. No LangChain to fight. Just nodes, edges, and results.
Start for free