Your coding assistant, your choice of AI
A multi-model AI chat built right into VS Code. Pick from 9 providers — Claude for complex reasoning, Ollama for local privacy, Gemini for million-token context — and switch anytime.
How the chat works
From prompt to applied code in three steps.
Pick a Provider
Open the MR Multi Model AI Coder sidebar. Click the provider chip to choose from 9 AI backends. Set your API key once — it's stored in your OS keychain. Ollama needs no key.
Ask Anything
Type your prompt, @-mention files, attach screenshots, or select code (Ctrl+Shift+L). Press Enter. The AI streams its response with syntax-highlighted code blocks.
Review & Apply
File edits show inline diffs — accept or reject per-change. Terminal commands show a confirmation. Read-only searches auto-approve. Rewind to any checkpoint if needed.
Core chat features
Not just a chatbot — an agentic coding partner that edits files, runs commands, searches your codebase, and manages tasks.
9 AI Providers
Claude, Kilo Code, Qwen, OpenRouter, Ollama, OpenAI Codex, Google Gemini, and HuggingFace. Pick the model that fits your task and budget.
Agentic Tool Loop
The AI can edit files (Edit / Write / MultiEdit), run terminal commands (Bash), search your codebase (Grep / Glob), and manage tasks (TodoWrite). Read-only tools auto-approve; edits and commands prompt for confirmation.
Inline Diff Preview
File edits show before/after at the line level with compressed unchanged regions. Accept or reject per-change. Open a full side-by-side VS Code diff editor if you prefer.
Streaming Responses
Responses stream token-by-token in real time. Stop anytime. Auto-scroll pauses when you scroll up, with a "Jump to latest" button for new content.
Multi-Tab Chats
Each session opens as a real VS Code editor tab. Run multiple conversations side by side. Past Conversations picker with date buckets and metadata.
Mid-Session Switching
Click the provider chip to switch models within a provider. Switching providers starts a fresh session with optional /fork summary to carry context.
Prompt History
Press Up/Down to cycle through previous prompts — terminal-style recall. Your draft is preserved. Per-message actions on hover: Copy, Edit & resend, Regenerate.
Markdown + Syntax Highlighting
Full Markdown rendering with syntax-highlighted code blocks and one-click copy. Tool messages collapse to a single line with click to expand. Long outputs show "Show more" past 24 lines.
@-Mentions for Context
Type @file to attach a file, @terminal:name to attach terminal output, @selection for editor selection. Live context readout shows input/output tokens, context %, and cost below the composer.
File & Screenshot Attachments
Paperclip button opens native OS file picker (multi-select). Shift+drop from VS Code explorer or OS. Paste screenshots with Ctrl+V — auto-saved and attached. Click chip to preview, x to remove.
Terminal Integration
The AI runs shell commands in your VS Code integrated terminal with full output visibility. Bash and BashOutput are separate tools — output is visible to both you and the AI.
Local-First with Ollama
Run models 100% on your machine. No API key, no network, no data leaving your computer. Availability probing with a 2-second timeout — the chip reflects live status.
Power features
Customize your workflow with permission modes, project conventions, scheduled messages, and more.
5 Permission Modes
Default (prompts for everything), Plan (read-only), Accept Edits (auto-approve file changes), Auto (auto-approve tools but AskUserQuestion still waits for you), Bypass Permissions (true full autonomy — auto-approves tools AND auto-answers questions with the recommended option). Switch anytime via slash command.
Checkpointing & Rewind
Every tool execution is checkpointed. Rewind to any previous state if the AI goes off-track. Session persistence stores everything locally.
Project Conventions (CODER.md)
Run /init to create a CODER.md with your coding rules — max lines, tech stack, testing policy, comment style, commit format, off-limits paths. Auto-loaded across all providers.
Scheduled Messages
Use /schedule to send a prompt at a future time. Get notified when the response is ready, even if VS Code is closed. Manage with /schedule list and /schedule cancel.
Haptic Feedback & Sounds
6 sound events: submit click, response ding, task complete chime, question tone, error thud, progress ambient. Customize with your own .wav/.mp3/.ogg files.
Slash Commands
/compact (compress context), /clear, /new, /fork <summary>, /cancel, /help, /cost (show session cost). Context compaction auto-scales to the target model's context window.
TodoWrite Live Checklist
When the AI uses TodoWrite, tasks render as a live checklist with in-progress and completed states. Track complex multi-step work visually.
Workspace Orientation
Auto-injected system prompt includes workspace root, detected project type, git branch, dirty file count, and open editor filenames. The AI starts every session with context.
Keyboard-first design
Every action has a shortcut. Stay in flow without reaching for the mouse.
| Shortcut | Action |
|---|---|
| Ctrl+Shift+L | Add editor selection to chat |
| Alt+K | Insert @path mention for current file |
| Ctrl+Escape | Toggle focus between editor and chat |
| Up / Down | Cycle through prompt history |
| Ctrl+V | Paste screenshot as attachment |
| Shift+Drop | Attach files from explorer or OS |
Free models you can use today
You don't need a paid API key to get started.
Ollama
Free foreverRun Qwen, Llama, DeepSeek locally. No API key, no network. Free forever.
Kilo Code
Free modelsFree models including Grok Code Fast (256K context) and Nemotron 120B via CLI.
Google Gemini
Free tierFree tier — Gemini 2.5 Flash (1M context, 500 req/day). No credit card required.
HuggingFace
Free tierOpen-weight models with free tier. Qwen3 235B, DeepSeek V3, Llama 4.
Start coding with AI in VS Code
Install the extension, pick a provider, and start your first conversation. Free models available — no credit card required.