Orchestrate every model on one canvas.
Futsu is a visual orchestrator for multi-LLM pipelines. Draw a graph, pin a model and instruction to each node, run it — and watch every step stream in real time.
●No credit card · Claude, Codex & OpenAI runners · Obsidian-native
Runs on the models your team already uses
Stop wiring LLM pipelines by hand.
YAML configs, glued scripts, and opaque chains don't scale. Futsu turns a multi-model pipeline into a graph you can see, run, and audit.
Visual, not YAML
Drag nodes, draw edges, branch and merge. The graph is the program — readable at a glance, diffable on the canvas.
Watch it run, live
Every node streams over PTY in real time. No black box — you see tokens, tools, and exit states as they happen.
Auditable by design
Runs persist as file-based artifacts under .futsu/runs/. Replay, inspect, and version every execution.
One canvas, deep capabilities.
Every piece has its own page — dive into the one you need.
Visual canvas
Draw the pipeline as a graph — nodes, bezier arrow edges, branch and merge on a pure-FSM engine.
Learn moreMulti-LLM runners
Claude, Codex and OpenAI on one pipeline, each step streamed live over PTY.
Learn moreToken credits
Buy tokens in bulk with automatic volume discounts. Spend them on any model — they never expire.
Learn moreObsidian bridge
Two-way sync with Obsidian JSON Canvas v1.0 — design a graph in either tool and stay in sync.
Learn moreAgents & Skills
Filesystem-based agents and skills from .claude/ — version them right alongside your repo.
Learn moreSecrets Vault
AES-encrypted secrets at user and workspace scope. Keys are hydrated at runtime, never on disk.
Learn moreThree steps to your first run.
Open a canvas
Spin up a project and drop your first nodes. No infra to configure — it runs locally.
Pin models & prompts
Attach a runner and instruction to each node, set parameters, connect the edges.
Run & iterate
Execute, watch the stream, inspect artifacts, refine. Repeat until it ships.
Serious tooling for people who ship with models.
Flat platform plans. Tokens by the bulk.
Platform access is a simple monthly fee — model usage is metered in tokens you buy in bulk.
Frequently asked.
Futsu ships with Claude, Codex, and OpenAI runners today, and you can pin a different model to every node in a single pipeline. Bring your own API keys — we never mark up tokens.
On the Solo plan everything runs locally and artifacts persist as plain files under .futsu/runs/. Cloud-hosted runs are an opt-in Pro feature.
Futsu's graph is the program. You compose, branch, and run visually on a pure-FSM engine, and watch execution stream in real time — no hand-maintained config files, no opaque chains.
Yes — Futsu has a two-way bridge with Obsidian JSON Canvas v1.0, so you can design a graph in either tool and keep both in sync.
Futsu is in active early access (Sprint-001). The canvas, multi-runner engine, and PTY streaming are live today — join early access to shape what ships next.
Your next pipeline starts on a canvas.
Draw it, pin your models, and run it in minutes. Free to start, no credit card.