Native sidebar
A chat-style panel sits in the editor. You prompt the model, watch the generated Flow DSL, review it, and run it without leaving VS Code.
Edition
Flow Code
The VS Code edition. A native sidebar where your selected local model generates a Flow DSL and the orchestrator runs it in-editor. The engine owns every file and shell action.
- Surface
- VS Code sidebar
- Execution locus
- Local machine (in VS Code)
- Status
- Shipped
The orchestrator, inside your editor
Flow Code drives the same engine as the desktop edition from a VS Code sidebar. You describe the work, review the generated flow, and the orchestrator executes it against your workspace under its own authority.
Generate, review, run
The selected local model writes a Flow DSL. You accept, revise, or reject it before anything runs. Manual mode prompts you on each fix, and autonomous mode applies them within budget.
Engine owns the actions
Every file write and shell command goes through the orchestration engine, not the model. The model proposes, and the engine executes and records.
In-editor monitoring
The model watches the run, and on an unhandled failure it proposes a corrected flow. The proposal is surfaced in the sidebar for review, or applied automatically in an autonomous turn.
Code intelligence built in
Flow Code builds a code-intelligence graph of your workspace from a fast native parser, with tree views for the graph and the blast radius of a symbol. You can search symbols, find callers and callees, and see what a change touches. When you review changed files, it assembles a focused, token-reduced context so the model reasons over what the edit affects.
Per-model config, shared settings
Each tab selects its own model, and a run-config popover sets the context size for local models and turns reasoning on or off. Settings live in one shared file that every edition reads, so the desktop app, server, and CLI all behave the same.
Editable runs you can export
The run panel shows the generated Flow DSL with syntax highlighting, and you can edit it in place before it runs. Export the DSL or the output whenever you need it, and a context-usage ring shows how much of the window the run is using.
Same core, same flows
Flow Code shares the Rust orchestration core and node catalog with every other edition, so flows move to and from the desktop and CLI unchanged.
JetBrains host (planned)
The same edition is planned for the JetBrains IDEs. It re-hosts the same engine sidecar and the same webviews in IntelliJ IDEA, PyCharm, WebStorm, GoLand, and the others, with only the editor-host glue rebuilt. VS Code is the shipped reference today.
Local by default
Inference runs on your machine, and operational data stays in your workspace. No third-party model service sees your code.
Add Flow Code to VS Code
Generate a flow from a prompt, review the graph, and run it in the editor. Inference stays local.