Open sourceMITLocal-firstv0.8.6
APIWeave turns your API calls into a visual workflow canvas. Chain requests, assert responses, and watch every run light up node by node — on your machine, with secrets that never leave it.
Windows · macOS · Linux — no account required
Real capture of the app — a workflow going green as it runs
Watch it flow
Press run and the canvas comes alive: requests fire along the edges, the camera follows the active branch, and each node settles as its assertions pass. When something breaks, it breaks loudly.
36 seconds: idle canvas → a node's request editor → a secret referenced by name → a token chained into the next request → the whole run going green. Captured from the desktop app.
A failing assertion paints its node red and routes the run down the failure branch. Open the node: the expected value, the actual response, and the operator that failed are right there — plus a waterfall of everything that already passed.
The workbench
Every visual here is the actual app. The colors are the app's own language: HTTP verbs, running amber, passing green, failing red.
Drag nodes onto the dot grid, wire them with bezier edges, and let the 700ms autosave do the saving. Workflows open as tabs; the minimap keeps big graphs navigable; JSON view is one click away.
Eleven operators over status, headers, cookies, variables and the previous response — with pass and fail edges, so a broken contract takes its own path through the graph.
An embedded terminal puts a coding agent beside the canvas, driving APIWeave over MCP: diagnose a failed run, patch the graph under a revision guard, and watch the change land live. It launches whichever CLI you already have installed.
Opt-in and loopback only: 52 whitelisted tools over 127.0.0.1, one bearer token per install, and a secret surface that returns names and scopes but never a value. Off by default — nothing listens until you ask.
The timeline lays every node out as a waterfall — parallel branches, delays and all. Open the failing bar for its reason, and the run's secret references stay masked.
JSONPath extractors chain responses into the next request, and the panel traces every value back to the node that produced it. Environments scope the values, and inheritance stacks them (base → staging → staging-eu).
Values are sealed into an encrypted store on your machine. Workflows reference {{secrets.*}}, the store is write-only at rest, and runs are masked before history is written.
Thirteen functions resolve at run time — uuid(), timestamps, dates, random strings and numbers. Search the panel, copy the placeholder, drop it into any value field.
Drop a specification, get a workflow
Import OpenAPI and Swagger definitions as ready-to-run request nodes, replay browser sessions from HAR, and paste a cURL command straight onto the canvas.
How it works
The loop the canvas is built around — the same three moves whether the workflow has five nodes or fifty.
Drag requests, assertions, delays and merges onto the canvas. Wire the branches, set expected statuses, extract what the next node needs. It autosaves as you go.
Pick an environment and press run. Requests fire along the edges, variables resolve, and the camera follows the active branch so you never lose the thread.
The timeline replays the run as a waterfall — every branch, every delay, every extraction, laid out on one clock. Open any node for its request, response and assertions.
Optional cloud
The desktop app is complete on its own. Add a free cloud account and your workflow structure follows you across machines — while everything sensitive stays exactly where it is.
The cloud API rejects any payload that carries secret material — a desktop client cannot upload one by accident.
Invite teammates, assign one of five roles, and share workspaces — with access you can revoke.
Every desktop that syncs is a known device. Revoke one, and it stops receiving structure.
When two machines change the same workflow, cloud raises a conflict — compare field by field and merge, never overwrite.
Cloud never runs your tests and never stores run history. It is a control plane, not a host.
Install the app, open the canvas, and watch your first workflow go green. No account, no configuration, no telemetry.
MIT · local-first · Windows, macOS, Linux