LangTailor Canvas
Design agent workflows in a visual canvas, save portable LangStitch IR documents, and export native Python, Java or Go projects.
LangTailor is available as a desktop IDE, a VS Code extension, and an Open VSX extension. Public installers and the canonical VSIX are hosted in langtailor-releases.
Version 0.4.0
- Native targets: Python with LangGraph and LangStitch SDK 0.3.2; Java with Spring AI; Go with a standard-library HTTP runtime.
- Portable documents: the shared spec 2.2.0 preserves graph state, target settings, asset configuration and native code bodies through migration and IDE round trips.
- Target-aware editing: Python, Java and Go bodies use their language editor and appear in generated source previews. Unsupported features produce explicit export errors.
- Actual toolchains: Build, Run and Test dispatch to Python, Maven or Go for the selected target. Process ownership keeps desktop windows from stopping one another's applications.
- Runtime behavior: supported exports use a common
/invoke request shape, API-key authentication, configurable host/port and HTTP tools. Python MCP invokes actual stdio, SSE and streamable HTTP servers.
- Marketplace: discover plugins, templates and integrations in the editor, and install supported assets from marketplace.langstitch.com.
Feature coverage differs by target. Native agent/RAG orchestration, MCP on Java/Go, durable checkpoints, streaming, debugging and reverse code synchronization are not all implemented. Rust is not an available production target. The platform panel reports unsupported configuration before export; it does not substitute an unrelated runtime.
Getting started
- Install LangTailor Canvas from the VS Code Marketplace or Open VSX, or download the desktop IDE.
- Create a graph with LangTailor: New Graph, or open a
*.langstitch.json document.
- Select the target in the platform panel, configure nodes and add code in the selected language.
- Export the project and follow its generated README. Build and Test require that target's local toolchain: Python 3.10+, Java 21 with Maven, or Go 1.22+.
For Python SDK installation:
pip install "langstitch-sdk[graph,server,llm,http,mcp]==0.3.2"
The separate Java IR compiler is available as com.langstitch:langstitch-spring-ai:0.2.2 on Maven Central. Go export is included in LangTailor; it is not a separate SDK package.
Documentation: sdk.langstitch.com.
Building from source
The canvas source lives in _canvas/. Clone the public shared spec beside this repository before installing the file dependency:
git clone --branch v2.2.0 https://github.com/LangStitch/langstitch-spec.git ../langstitch-spec
npm ci --prefix ../langstitch-spec
npm run build --prefix ../langstitch-spec
npm ci
npm run build
npm run package
The VSIX is written to dist-langtailor/langtailor-canvas-0.4.0.vsix. For desktop development, install _canvas and desktop dependencies, then run npm run build --prefix desktop.
Releases
CI executes generated Python, Java and Go applications across Windows, macOS and Linux, builds the extension, and compiles desktop sources. Release jobs build the Windows x64 installer and macOS x64/arm64 DMG and ZIP files.
Set the matching version in the extension and desktop manifests and both lockfiles, then push langtailor-v<version>. The release workflow validates versions and native conformance, checks updater hashes and the complete artifact set, publishes the public GitHub release, and sends that exact VSIX to both extension stores. Individual store workflows can retry publication from an existing release without rebuilding or replacing it.
Each release includes SHA256SUMS, release-manifest.json, updater manifests and blockmaps. These checks establish artifact integrity and build provenance; they do not certify every production deployment or installer interaction. Platform signing must be configured separately. Current unsigned desktop releases may trigger Windows or macOS security prompts; macOS automatic update installation requires signing.
License
MIT © LangStitch
