Everything you need to build and run AI workflows
Build typed AI pipelines visually or from an agent, then run them anywhere, from the API, SDK, MCP, or CLI.
- HTTP API
- TypeScript SDK
- MCP server
- CLI
Design typed graphs, visually
Drag nodes onto an infinite canvas and wire them together. Group related nodes into frames, drop in notes, snap to a grid, undo anything, and auto-layout the whole graph left to right. Every change saves as you work.
One workflow, every surface
Publish once and call the exact same workflow from a REST API, the TypeScript SDK, the MCP server, or the CLI. Every surface is generated from a single typed contract, so inputs, outputs, and errors match wherever you invoke it, and results stream back live.
Reproducible, down to the version
Publishing freezes a workflow into an immutable version and pins every node it uses to an exact release. Pin @2.1.0 for a build that never changes, @2 to ride the latest patch, or @latest to always get the newest. Your choice, per run.
- flux-schnellv2.1.0
- recraft-v3v1.0.0
- constv1.2.0
A complete platform, not just an editor
From the first node to production traffic: building, publishing, running, and operating all live in one place.
Build with an agent
Describe the pipeline you want and an in-editor agent assembles the graph, adding nodes, wiring ports, and exposing inputs. Refine by hand or keep prompting; the result is the same typed workflow.
Type-safe by construction
Ten built-in connector types (image, text, audio, video, floats, ints, booleans, embeddings, files, and structured data), plus your own. Ports only connect when the types line up, so a graph is correct before it runs.
Publish & reuse
Freeze a workflow under an org/name with a semver bump, then drop it into other graphs as a single node. Roll back to any prior version from the history at any time.
Metered to the step
Runs draw on a prepaid balance and are charged exactly what they meter, at published per-node rates. Each step reports its cost live as it completes, in the editor, the API, and the CLI alike.
Observability built in
A per-org dashboard of runs and spend: cost over time, top workflows and nodes, and recent runs, filterable by API key, refreshing as runs land.
Teams & roles
Organizations with members, roles, and email invitations. Role-based permissions gate who can edit workflows, change roles, or manage keys, and canvas comments keep feedback in place.
Call it from anywhere
Every surface is generated from one typed contract, so they stay in lockstep, with the same inputs, outputs, and errors on each.
HTTP API
REST endpoints under /v1, authenticated with a bearer key. Streams run events over SSE and ships an OpenAPI spec generated from the contract.
TypeScript SDK
@blitflow/sdk gives you run() to await outputs or startRun() to iterate live events, fully typed end to end, validated on the way in and out.
MCP server
A hosted, OAuth-protected server at mcp.blitflow.com lets agents run your workflows as tools, plus a local stdio server for your own machine.
CLI
blitflow run a workflow file or a published ref, watch progress stream in your terminal, and check spend with blitflow usage.
A curated library of typed nodes
Image and language models, media utilities, and operators, all behind one uniform, typed node interface, so every node wires up and runs the same way.
Image models
53 nodes- flux-schnell
- flux-dev
- flux-kontext-dev
- ideogram-v3-turbo
- recraft-v3
- nano-banana-2
- gpt-image-2
- real-esrgan
Language model
text & structured output- claude-sonnet-4-6
- gpt-5-4
- gemini-3-flash
Audio
transcription & speech- whisper
- kokoro
Media utilities
6 nodes- image-resize
- image-format-convert
- video-to-gif
- extract-frame
- audio-convert
- background-remover
Build one pipeline. Ship it everywhere.
Start in the studio for free. No card, no setup. Publish when you're ready and call your workflow from anywhere.