Give intentsolid ground.

The semantic layer between agents and real software. Discover capabilities, control authority and inspect the outcome.

Name the operation.

A typed request addresses an application object.

Start with Semwright.

One runtime for the CLI, MCP and inspector.

Linux installation

Download aarch6415,564,681 bytesSHA-256 1c0cfa81ee53fafa029208b6b60442cc8923838428a12ccdd9f745089fff73d8
Download x86_6416,711,910 bytesSHA-256 2b927289b949b2b0b927d10869f5a845b7581a538eee829ef3495e2ba268c385

Verify the manifest before extracting. Then run ./install.sh from the extracted directory, then runsemwright setup using the exact installed path printed by the helper. Copy the generated MCP snippet into your client.

macOS installation

Download arm6412,728,295 bytesSHA-256 20b6831670151a3bdf55a22324810a15213bb4288af5d9e7221961fd7318cb07
Download x86_6413,712,538 bytesSHA-256 57511ac1dee3140438ffa580db91bbc0de545735ccd07f0b65c8493ef7fbb697

Verify the manifest before extracting. Then run ./install.sh from the extracted directory, then runsemwright setup using the exact installed path printed by the helper. Copy the generated MCP snippet into your client.

Windows installation

Download arm6413,346,436 bytesSHA-256 d6acb085d5e7cf33ee27d5179b59e2e958cde10d3260df28c0e8649632a43089
Download x86_6414,542,169 bytesSHA-256 7292e273ff7bd0e21598f5c5b085315b1bc7766f0d5f68da9efb22626d53308b

Verify the manifest before extracting. Then run .\Install-Semwright.ps1 from the extracted directory, then runsemwright setup using the exact installed path printed by the helper. Copy the generated MCP snippet into your client.

Version 1.0.0. Source 8fa19125.

Independent security review is incomplete. Read the publication exception.

One prompt. A connected creative workflow.

Watch Semwright update an aircraft’s materials in Blender, render a new camera shot, generate branded titles in Motion Canvas, and assemble the video with audio through MLT—all using Semwright’s drivers.

Recorded in one take using an existing aircraft project. Waiting periods are accelerated; the final playback runs at normal speed. The workflow produces a new video and an editable timeline.
Watch the 60-second demo.

Software has structure.
Agents should use it.

Work with a document’s nodes, a scene’s objects or a timeline’s clips. Semwright turns application meaning into an inspectable contract.

Semantic objects

Named objects and typed capabilities, exposed through application interfaces.

Policy boundaries

Authority evaluated before execution, with explicit scope and limits.

Refs and jobs

Object identity and ongoing work with lifetimes you can inspect.

Observable results

Read back state, evaluate effects and keep the execution evidence.

A clear path from
intent to action.

Each boundary has a job. Each operation leaves something you can inspect.

AgentTyped intent

A named operation with explicit inputs.

Semwright broker
Capability registry

Schemas, permissions and execution contracts.

Application driversNative operations
  • Figma and browser
  • Blender and Godot
  • Media and documents
Readback and audit

Authority before action. Discovery does not grant execution. Policy evaluates the requested operation and its scope.

Identity with a lifetime. References address objects within a session. Stale or invalid references must be resolved again.

Work you can inspect. Session-bound jobs expose progress and results, with cooperative cancellation where supported.

Interactive architecture illustration. No application is connected. Read the execution model.

Different tools.
Shared meaning.

Each driver describes its application’s objects, operations and limits.

Explore all drivers

Figma

Address pages, nodes, components, variables, prototypes and exports through the official Figma Plugin API.

Experimental

Chromium

Work with tabs and the DOM, not an estimate of where a button might be.

Experimental

Blender

Reach into the scene through an application-native driver.

Experimental

MLT video

Address timelines, tracks and clips as structured objects.

Experimental

Inside the application.
Beyond the pixels.

A frame, a scene object, a material. Each has an identity and a place in its application's model. Work with the things the software actually knows.

Explore native drivers
Inside the documentObject study
FrameLandscape

An image inside a frame, with bounds and a place in the document.

Interactive object illustration. No application is connected.
Inside the sceneBlender study
cobalt
ObjectOrbit mesh

Geometry with a name, a transform and a place in the scene.

Interactive Blender object illustration. No application is connected.

Frame

Layout, hierarchy and geometry.

Component

Reusable structure and instances.

Variable

Values, modes and bindings.

Inspect the real Figma proof

Real work in real applications.

A game built across Blender and Godot. A light study rendered in Godot. Watch the results, then explore the work and its evidence.

Make a workflow
worth repeating.

Discover an operation. Resolve its target. Request an action within policy. Keep the same execution boundary as a workflow grows.

Explore workflow distillation
Semwright RequestRequest illustration

Ask for the thing you mean.

The agent calls a typed operation instead of guessing a screen coordinate.

Operation
ui.find
Application
Text Editor
Looking for
Save button
No pixel guessing.Shared execution path

Your agent gets access.
You keep the keys.

Discovering a capability does not grant authority to run it. Input, capture and application changes pass through explicit policy.

See how permissions work
Semwright ControlPolicy illustration

Choose what an agent may do.

Discovery can stay broad while sensitive actions remain separately controlled.

InputKeyboard and pointerGrant required
ClipboardRead or write clipboard dataGrant required
CaptureScreen and screenshot accessGrant required
App changesNative write or destructive operationsPolicy checked

Build with the
whole picture.

From your first safe command to an application-owned integration. Follow the part of the system you need.

Open the documentation

Before you connect.

Installation, permissions and platform scope, with the current limits in view.

Is Semwright another AI agent?

No. Semwright is the interface an agent calls, not the model that decides what to do. It exposes capabilities and mediates execution. You bring the agent.

Why not just give an agent a shell?

A shell usually inherits broad user authority. Semwright is built around narrower operations, scoped references, explicit permissions and one audit path.

Is this only an MCP server?

MCP is one frontend. The CLI, inspector and recipes use the same broker. Application drivers and federated MCP providers extend the capability catalogue.

What can I use today?

Semwright 1.0.0 native bundles include the CLI, daemon, MCP frontend and inspector. Install, run semwright setup and connect your client. Application coverage stays version-specific; read the compatibility guide for the evidence and limits.

What does cross-platform mean today?

Native bundles are available for Linux, macOS and Windows on x64 and ARM64. Packaging tests and portable SDK tests are distinct from physical desktop acceptance. Read the compatibility guide for the environment-specific limits and incomplete security-review status.

Meaning, made
programmable.

Bring your agent. Start with the source. Follow the evidence.

Source snapshot 2026-10-07: 8fa19125. Historical demonstrations retain their own source revisions. Read the scope and limits.