Kyrex for VS Code

A local Python engine, with your choice of model provider.

Kyrex runs its coding tools in a Python subprocess in your workspace. Chat, active-file context and relevant tool results are sent to your configured model provider. Choose a local provider such as Ollama to keep model inference on your machine. Remote providers receive the context needed for their responses.

Key Features

Active file context — Every message automatically includes the content of your active editor tab. Kyrex always knows what you're looking at.
Streaming responses — Token-by-token streaming in the sidebar. Watch reasoning unfold in real time.
Tool visibility — Every tool call is displayed inline with expandable results. See exactly what Kyrex is doing and why.
Diff gate — Full file rewrites open a native VS Code diff view for review and approval. Surgical edits apply directly with a visible inline diff.
Session management — Token tracking, model switching, and one-click session resets. Start fresh or keep context — your call.

Requirements

VS Code 1.85 or later
Python 3.10+ (the Kyrex engine runs as a local Python subprocess)
API key for any OpenAI-compatible provider (OpenAI, OpenRouter, Ollama, LM Studio, vLLM, etc.) or Anthropic

Installation

Search for Kyrex in the Extensions panel.
Ensure Python 3.10+ is available on your PATH.
Install dependencies into the interpreter selected by kyrex.pythonPath: python -m pip install openai anthropic requests textual. On Windows, use python or the full interpreter path if python3 is unavailable. In WSL/SSH, install into the interpreter on that host.
Open the sidebar Settings, select a provider, enter its API key and paste a model ID. Click Apply settings once. Wait for Engine ready.

To install a prepared update, use Extensions → … → Install from VSIX, select the .vsix, then run Developer: Reload Window. Check the installed version. Marketplace publication is a separate release step.

Configuration

All settings are under kyrex.* in your VS Code settings (Ctrl+, → search "Kyrex"):

Setting	Description	Default
`kyrex.provider`	LLM provider (`openai` or `anthropic`)	`openai`
`kyrex.model`	Model name (e.g. `gpt-4o`, `claude-sonnet-4-5-20250929`)	(empty — uses provider default)
`kyrex.apiKey`	Legacy setting; global values migrate to SecretStorage	(empty)
`kyrex.baseUrl`	Custom API endpoint URL	(empty — uses provider default)
`kyrex.pythonPath`	Path to the Python interpreter	`python3`

Sidebar keys are stored in VS Code SecretStorage and saved keys are never returned to the sidebar. Leave the key input blank to keep the saved key. Clear saved key takes effect on Apply settings; environment keys may still be used. Existing workspace-specific legacy keys remain workspace-specific. You can also supply an environment key when launching VS Code:

export KYREX_API_KEY="sk-..."

Usage

Open the sidebar — Click the Kyrex icon in the Activity Bar.
Ask anything — Type a question and press Enter. Your active file is sent as context automatically.
Review tool calls — Expand any tool call to inspect arguments and results.
Approve edits — When Kyrex proposes a file change, review the diff and click Apply or Reject.
Switch models — Open the Settings panel at the bottom of the sidebar to select a provider or paste a model ID, then click Apply settings. Typing does not restart the engine.
New session — Click + New to clear context and start fresh.

Supported Providers

Kyrex works with any OpenAI-compatible API out of the box:

Provider	Base URL	Notes
OpenAI	(default)	Set `kyrex.provider` to `openai`
Anthropic	(default)	Set `kyrex.provider` to `anthropic`
OpenRouter	`https://openrouter.ai/api/v1`	Hundreds of models, single key
OpenCode Go	`https://opencode.ai/zen/go/v1`	Paste your model ID
Ollama	`http://localhost:11434/v1`	Fully local, no API key needed
LM Studio	`http://localhost:1234/v1`	Local models with OpenAI-compatible API
vLLM	`http://localhost:8000/v1`	High-throughput local inference
Any OpenAI-compatible	Your endpoint	Set `kyrex.baseUrl` to your server

Set a custom API endpoint and key in the sidebar and apply them together. Opening Settings reads the catalogue from the saved endpoint. If it is unavailable, paste a model ID directly. Unapplied drafts are not sent.

Commands

Type /usage in the sidebar to see input/output token totals and an available cost estimate for the session. Counts include repeated context across model calls and may include estimates; check your provider for billed usage. Automatic usage updates are hidden to keep the header clear.

To recover a stalled engine, expand Settings and click Restart engine.

Command	Description
`Kyrex: Start Engine`	Manually start the Python engine
`Kyrex: Stop Engine`	Stop the engine process
`Kyrex: Send Message`	Programmatically send a message

Building an update

From kyrex-vscode:

npm ci
npm run lint
npm run build
npm test
npm run package
npm run verify:package

Build, watch startup and VSIX packaging generate a fresh kyrex_engine directory from the shared source. Do not edit the generated directory. Verification compares every packaged runtime file with its source and excludes tests, caches, dependencies and local sessions. CI builds packages on Linux and Windows. The VSIX contains engine source and assets; install Python dependencies separately.

Disclaimer

Kyrex modifies files on your system. Always use version control. The authors are not responsible for any data loss or damages resulting from use of this software.

License

MIT