MiMo for Copilot Chat

Bring MiMo, DeepSeek V4, Chat Completions, and OpenAI Responses-compatible models into the native GitHub Copilot Chat experience.

Install from Marketplace · Quick Start · Configuration

⚡ Quick start

1) Open setup from the Command Palette

• Press Ctrl+Shift+P on Windows/Linux or Cmd+Shift+P on macOS.
• Type one of these commands:
  • MiMo: Set API Key
  • MiMo: Open Provider Configuration

This is the fastest way to configure the extension after installation.

2) Pick a model in Copilot Chat

After configuring a provider, open Copilot Chat and use the model picker.

• You can choose MiMo / DeepSeek / Responses-compatible models directly in the picker.
• You can also use Manage Models... from the picker when needed.

3) Find the configuration page from the token indicator

After the extension is active, look at the bottom-right status bar in VS Code.

• A MiMo token usage indicator appears there.
• Click it to open the Provider & Model Configuration page.
• Its tooltip shows the real API prompt-token value after MiMo preprocessing. When MiMo compresses tool output, it also shows the estimated pre-compression value and compression ratio.

This is the easiest always-visible entry point once the extension is running.

✨ Overview

MiMo for Copilot Chat plugs additional provider-backed models directly into the Copilot Chat model picker. It keeps the built-in Copilot workflow intact while adding support for:

• 🧠 Xiaomi MiMo models
• 🔷 DeepSeek V4 models
• 🌐 Chat Completions and OpenAI Responses-compatible providers
• 🔑 provider-specific API keys and Base URLs
• 👁️ model visibility controls
• 🧩 memory-mode configuration

🗺️ Visual at a glance

flowchart LR
  A[🔑 API Key] --> B[⚙️ Provider Configuration]
  B --> C[🧠 Copilot Chat Model Picker]
  C --> D[💬 Chat / Agent Mode]
  B --> E[🧠 Memory Mode]
  B --> F[👁️ Model Visibility]
  G[MiMo / DeepSeek / Responses] --> B

🎯 What you get

• 🚀 Use new models inside the native Copilot Chat UI
• 🛠️ Keep agent mode, tool calling, instructions, workspace search, and editor actions
• 🔒 Store keys securely in VS Code SecretStorage
• 🔁 Switch between providers without changing your workflow
• 🧰 Configure built-in and custom models from one place

⚙️ Configuration page

The configuration page includes:

• API provider management
• model management
• Agentic Memory settings
• Responses tool-output compression settings

The configuration UI is adapted from the OAI-compatible Copilot-style settings experience and customized for MiMo providers, model routing, memory, and Responses compression controls.

Provider API mode

When editing a provider, choose the API mode that matches the endpoint and then click Save:

• Chat Completions for OpenAI-compatible /chat/completions providers, for example a base URL like https://api.anyone.ai/v1.
• Responses for providers that expose the OpenAI /responses API.

If a provider was created with the wrong mode, open Provider Configuration, click Edit, change API Mode to Chat Completions or Responses, and save it again.

Agentic Memory

Use this section to enable memory recall and choose the memory model.

Responses tool-output compression

Use this section to control:

• real image compression
• format conversion before resize
• long-output truncation
• tool-type policies
• structured summaries
• replay behavior for original images

If a tool image remains too large and has to be removed, MiMo always emits a bold chat warning even when optional chat notices are disabled.

🧪 Supported providers

• mimo → https://api.xiaomimimo.com/v1
• mimo-tp → https://token-plan-cn.xiaomimimo.com/v1
• deepseek → https://api.deepseek.com
• openai-responses → any compatible /responses endpoint

🧬 Supported models

• MiMo V2.5 Pro
• MiMo V2.5
• MiMo V2 Pro
• MiMo V2 Flash
• DeepSeek V4 Pro
• DeepSeek V4 Flash
• GPT-5.4
• GPT-5.5 — defaults to 258K input / 128K output tokens and can be overridden in model settings

📦 Install

🛒 Marketplace

Install the extension from the VS Code Marketplace.

📎 VSIX

1. Download or build the VSIX.
2. Run Extensions: Install from VSIX... in VS Code.
3. Select the .vsix file and reload the window.

🧭 Typical workflow

1. Open the Command Palette with Ctrl+Shift+P / Cmd+Shift+P.
2. Run MiMo: Set API Key or MiMo: Open Provider Configuration.
3. Add one or more providers.
4. Open Copilot Chat and pick a MiMo / DeepSeek / Responses model.
5. Use the bottom-right token usage indicator whenever you want to reopen the configuration page.
6. Optionally configure memory mode and Responses tool-output compression.

🧰 Configuration

Key settings include:

• mimo-copilot.providers
• mimo-copilot.models
• mimo-copilot.hiddenModels
• mimo-copilot.maxTokens
• mimo-copilot.agenticMemory
• mimo-copilot.memory.recallModel
• mimo-copilot.modelIdOverrides

📝 Notes

• Copilot Chat features remain available.
• DeepSeek and MiMo routing depend on the configured provider and API key.
• API keys are not written to settings.json.
• The status-bar token indicator is the recommended entry point for reopening the configuration page.

🛠️ Development

npm install
npm run compile
npx vsce package

🙏 Fork & Credits

This project is forked from and based on Vizards/deepseek-v4-for-copilot.

Thanks to the original author and contributors for the DeepSeek Copilot Chat provider foundation. This fork keeps the upstream MIT license notice and extends the project with MiMo provider support, multi-provider configuration, memory features, and Responses-compatible model support.

License

MIT