Drag agents, tools and connectors onto a canvas, wire them together, and watch every node execute live. Automate with schedules and webhooks, keep humans in the loop, and analyze everything your workflows produce — all in one studio.
No sign-up needed — the demo is open. Powered by GenXAI, a framework for multi-agent AI.
Flow Studio pairs an n8n-style visual builder with real AI agents, durable data, and the operational tooling to run workflows for real.
Drag & wire inputs, agents, tools, decisions, loops and subworkflows. Live per-node status streams onto the canvas as runs execute.
Attach a Chat Model, persistent Memory, and any tools to an agent's ports — the LLM calls them mid-reasoning. Team patterns wire multi-agent collaboration.
Type what you want — the studio's generator drafts the whole workflow on the canvas, and can refine the one you're editing.
Run on a schedule (intervals or cron with timezones) or fire from webhooks — including signed GitHub events and WhatsApp Business messages. Toggle triggers on/off with one click.
Approval gates pause a run until someone responds. Expressive branch conditions route on any node's output. Failed runs retry from the failed node.
Ten built-in connectors, 46 agent tools, and any Model Context Protocol server — every connector action doubles as an agent tool.
Workflows accumulate rows into durable datasets. Query federated sources, and let agents build dashboard reports — charts plus narrative, downloadable as HTML.
State an objective and a multi-agent pipeline plans, explores, cleans, engineers features, trains models and reports — with code review and optional human gates at every stage.
KPIs, runs-per-day, trigger breakdowns, slowest nodes and a live activity feed — auto-refreshing while your automations run.
Drag nodes from the palette — or describe the workflow and let the generator draft it. Configure each node in the inspector; pass data between nodes with {{ }} expressions.
Hit Run and watch every node light up with streamed status. Click any node to inspect its output. Versions snapshot on every save, restorable anytime.
Add a trigger — schedule or webhook — and the workflow runs itself. Track everything in the ops dashboard, and route failures to an error-handler workflow.
Plus 46 built-in agent tools (HTTP & API calls, web scraping, PDF/CSV/JSON/Excel processing, SQL, vector search, code execution, dataset queries, …) and any MCP server you register — and every connector action above doubles as an agent tool.
Flow Studio is the canvas — GenXAI is the machine. A production-grade framework for multi-agent AI where agents don't just answer: they collaborate, remember, delegate, disagree and decide. Every workflow you draw compiles onto its graph engine — the same engine you can drive from code.
Conditional branches, parallel fan-out with real join semantics, loops, nested subworkflows, checkpoints and resume. Not a task queue — a living graph.
Coordinator/workers, round-robin relays, map-reduce swarms, critic loops and voting councils — collaboration patterns you compose, not hard-code.
Agents recall past runs, share a common memory bus mid-flight, and pick up conversations where they left off — days later.
45 built-in tools, every connector action, and any MCP server — handed to agents so they can act mid-reasoning, with human gates where it matters.
Cron and interval schedules, signed webhooks, event triggers — workflows that run because the world changed, not because someone clicked.
Every workflow, node, agent and tool call is metered — Prometheus metrics out of the box, run history down to per-node timings.
Everything the canvas does is a library call away. Define agents in code, run workflows from YAML, or mix both — the studio and the SDK share one engine.
Self-hosted, container-ready, Kubernetes-friendly. Your models, your data, your infrastructure — the framework scales from a laptop to a cluster.
The public demo at app.genxflowstudio.com is open — no sign-up needed — and the Studio is self-hostable on your own infrastructure. It's built on the MIT-licensed open-source GenXAI framework.
No. Drag nodes onto the canvas, configure them in the inspector, and pass data with simple {{ }} expressions — or describe the workflow and let the generator draft it. Everything the UI does is also available via the REST API.
Ten built-in connectors (Email, Slack, WhatsApp, HubSpot, S3, PostgreSQL, GitHub, Jira, Notion, Google Workspace), 46 built-in agent tools, and any MCP server you register — every connector action doubles as an agent tool.
Yes — a FastAPI backend with PostgreSQL persistence, shipped as a single container and Kubernetes-ready. Your models, your data, your infrastructure.
Per-node retries with backoff, continue-on-fail so downstream nodes can react, retry-from-failed-node, and routing to an error-handler workflow. Runs are durable — closing the browser doesn't kill them.
The open-source GenXAI framework: every workflow you draw compiles onto its graph engine — the same engine you can drive from Python. Explore it at genxai-framework.com.
The public demo ships with example workflows — chained tools, agent teams, approval gates, dataset collection. Open one and press Run.
Open the Studio →