Getting Started

Tip: the Get Started with SideCar walkthrough auto-opens in VS Code’s Welcome editor the first time you install SideCar. It walks through the five essentials in under a minute and auto-checks off each step as you run the relevant command. Reopen it any time with SideCar: Open Walkthrough from the Command Palette.

Install Ollama

Download and install Ollama for your platform. After installation, verify it’s working:

ollama --version

Pull a recommended model:

ollama pull qwen3-coder:30b

Install SideCar

Install from the VS Code Marketplace, or search for “SideCar” in the VS Code Extensions panel (Cmd+Shift+X / Ctrl+Shift+X).

Your first chat

1. Click the SideCar icon in the activity bar (left sidebar) or press ⌘⇧I (Ctrl+Shift+I on Windows/Linux) to toggle the chat panel.
2. The empty-state welcome card shows you the active model, three quick-action buttons, and four starter prompts you can click to pre-fill the input.
3. Type a message in the chat input or click a starter prompt.
4. SideCar auto-launches Ollama if it’s not already running.

SideCar will automatically include your active file and workspace context in the conversation.

Discover every action

Press ⌘⇧P (Ctrl+Shift+P on Windows/Linux) and type SideCar: — every user-facing action is in the palette with a consistent prefix and icon. Useful ones to remember:

• SideCar: Toggle Chat — open or hide the chat sidebar
• SideCar: Inline Chat — inline edit prompt in the editor (⌘I)
• SideCar: Select Model — keyboard-first model picker
• SideCar: Set / Refresh API Key — paste or rotate your key
• SideCar: Switch Backend — flip between Ollama / Anthropic / OpenAI / Kickstand
• SideCar: Show Session Spend — $ breakdown for the current session on paid backends
• SideCar: Open Walkthrough — reopen the getting-started page

Code actions on diagnostics — when VS Code shows a red or yellow squiggle on a line, press ⌘. / Ctrl+. and you’ll see Fix with SideCar and Explain this error with SideCar alongside the built-in Quick Fix suggestions. Refactor with SideCar appears in the Refactor submenu on any selection.

Choosing a model

Use the model dropdown at the top of the chat panel to browse and switch models. Models are organized into two categories:

• Full Features (Tools) — models that support function calling (e.g., qwen3-coder, llama3.1, command-r)
• Chat-Only — models that work for conversation but can’t use tools (e.g., gemma2, llama2, mistral)

You can also install new models directly from the dropdown — SideCar will pull them from the Ollama registry.

The default model is qwen3-coder:30b. For machines with less RAM, try qwen3-coder:8b or qwen2.5-coder:7b.

Switching backends

The fastest way to move between Ollama, Anthropic, OpenAI, and Kickstand is the ⚙ gear button in the chat header. It opens a settings menu with a Backend section — pick a profile and SideCar flips baseUrl, provider, and model in one click. Each profile keeps its own API key in VS Code’s SecretStorage, so switching doesn’t clobber keys you’ve already set. The same flow is available from the Command Palette as SideCar: Switch Backend.

The sections below describe each backend in detail and also cover the manual settings path if you prefer editing settings.json.

Using the Anthropic API

Recommended: click the ⚙ gear → Anthropic Claude. SideCar prompts for your API key on first switch, saves it to the Anthropic-specific SecretStorage slot, and sets baseUrl / provider / model for you.

Manual:

1. Set sidecar.baseUrl to https://api.anthropic.com
2. Run SideCar: Set / Refresh API Key from the command palette and paste your Anthropic API key
3. Set sidecar.model to a Claude model (e.g., claude-sonnet-4-6)

SideCar uses prompt caching with Anthropic, reducing input token costs by ~90% on cache hits.

Note: the Anthropic API is a separate paid service from Claude.ai subscriptions — your Max or Pro plan does not include API credits. Get a key at platform.claude.com.

Using Kickstand

Kickstand is a self-hosted local inference server that manages model loading, unloading, and GPU memory efficiently. CLI command: kick.

1. Start the server with kick start.
2. Recommended: click the ⚙ gear → Kickstand in the chat header. Manual: set sidecar.baseUrl to http://localhost:11435 (default Kickstand port).
3. Set sidecar.model to the model you want to use.

SideCar auto-detects Kickstand by the port number and reads the bearer token from ~/.config/kickstand/token automatically — no API key prompt or settings plumbing needed.

Using OpenAI-compatible servers

SideCar works with any server that exposes the OpenAI /v1/chat/completions endpoint — including LM Studio, vLLM, llama.cpp, text-generation-webui, and OpenRouter.

1. Set sidecar.baseUrl to your server’s URL (e.g., http://localhost:1234)
2. Set sidecar.apiKey if required (optional for most local servers)
3. Set sidecar.model to the model name your server is running

SideCar auto-detects the provider from the URL. If auto-detection gets it wrong, set sidecar.provider explicitly:

"sidecar.provider": "openai"

Provider examples

Tool support

Most OpenAI-compatible models support function calling (tool use), which enables SideCar’s full agentic capabilities — file editing, shell commands, git operations, etc. If a model doesn’t support tools, SideCar falls back to chat-only mode automatically.

Keyboard shortcuts