+ Providers on Copilot Chat
Connect LM Studio, Ollama, and OpenCode to GitHub Copilot Chat — local and remote AI models
📥 Download VSIX
·
GitHub
·
Report Bug
VS Code extension that registers LM Studio, Ollama, and OpenCode as Language Model providers for GitHub Copilot Chat. Use your own local or remote AI models directly in Copilot Chat.
📁 Architecture
src/
├── providers/
│ ├── BaseProvider.ts # Base class for all providers
│ ├── sdk/
│ │ ├── anthropicChat.ts # Anthropic SDK handler
│ │ ├── openaiChat.ts # OpenAI SDK handler
│ │ └── utils.ts # Shared message conversion utilities
│ ├── LMStudioProvider.ts # LM Studio (OpenAI-compatible API)
│ ├── OllamaProvider.ts # Ollama (OpenAI-compatible API)
│ ├── OpenCodeFreeProvider.ts # OpenCode Free tier
│ ├── OpenCodeGoProvider.ts # OpenCode Go tier
│ ├── OpenCodeServerProvider.ts # OpenCode Server
│ └── OpenCodeZenProvider.ts # OpenCode Zen tier
├── client/ # API clients
├── config/ # Settings and storage
└── ...
Key Design Decisions:
- All providers extend
BaseProvider (formerly OpenAICompatibleProvider)
- SDK-based handlers (
anthropicChat.ts, openaiChat.ts) use official AI SDK packages
- Shared utilities in
sdk/utils.ts eliminate code duplication
- LM Studio and Ollama use OpenAI-compatible API (not custom SSE parsing)
✨ Supported Providers
| Provider |
Connection |
Models |
Features |
| LM Studio |
Local or Remote |
Auto-detected from /v1/models |
Streaming, Reasoning, Vision, Tools |
| Ollama |
Local or Remote |
Auto-detected from /api/tags |
Streaming, Reasoning, Vision, Tools |
| OpenCode Free |
Cloud |
4+ |
DeepSeek V4 Flash Free, MiMo V2.5 Free, Nemotron 3 Super Free |
| OpenCode Go |
Cloud |
16+ |
Kimi K2.5/K2.6, DeepSeek V4 Pro/Flash, GLM 5/5.1, MiMo, MiniMax, Qwen |
| OpenCode Zen |
Cloud |
66+ |
GPT 5.x, Claude Opus/Sonnet/Haiku, Gemini 3.x, Kimi, DeepSeek, GLM, Grok |
| OpenCode Servers |
Local or Remote |
∞ |
Any model configured in your OpenCode server |
🔌 How to Connect
LM Studio (Local or Remote)
- Install LM Studio and start the local server (default port
1234)
- Or use a remote LM Studio instance
- The extension auto-detects all loaded models with their capabilities
Configuration: settings.json
{
"lmstudio.baseUrl": "http://localhost:1234"
}
Ollama (Local or Remote)
- Install Ollama and pull models:
ollama pull llama3.1
- Start Ollama server (default port
11434)
- Or use a remote Ollama instance
- The extension auto-detects all pulled models with their capabilities
Configuration: settings.json
{
"ollama.baseUrl": "http://localhost:11434"
}
OpenCode Servers (Local or Remote)
- Install OpenCode CLI:
npm install -g opencode
- Start server:
opencode serve (default port 4096)
- Or connect to a remote OpenCode server
- Add server via command palette:
+ Providers: Add Server
Features:
- Multiple servers simultaneously
- Basic auth support (
OPENCODE_SERVER_PASSWORD)
- Session-based API:
POST /session + POST /session/:id/message
OpenCode Cloud (Free/Go/Zen)
- Get API key at opencode.ai/auth
- Configure via command palette:
+ Providers: Configure Zen/Go/Free
🧠 Model Capabilities Detection
The extension automatically detects model capabilities:
- Reasoning: DeepSeek, Qwen3, models with "reasoning" or "think" in name
- Vision: LLaVA, models with "vision" or "vl" in name
- Tools: Qwen, Llama3, models with "tool" or "function" in name
- Context Length: Heuristic based on model size (7B→8K, 13B→32K, 70B→128K)
🌐 Remote Servers
All local providers (LM Studio, Ollama, OpenCode Servers) support remote connections:
{
"lmstudio.baseUrl": "http://192.168.1.100:1234",
"ollama.baseUrl": "http://my-server:11434"
}
📦 Installation
From VS Code Marketplace
Search for "+ Providers on Copilot Chat" in the Extensions panel.
From VSIX
- Download
.vsix from Releases
- Run:
code --install-extension plus-providers-copilot-chat-3.3.0.vsix
🛠️ Requirements
- VS Code 1.120.0+
- GitHub Copilot Chat extension
- For local models: LM Studio, Ollama, or OpenCode CLI running
📝 Changelog
See CHANGELOG.md for version history.
📦 Installation
From VS Code Marketplace
Search for "+ Providers on Copilot Chat" in the Extensions panel.
From VSIX
- Download
.vsix from Releases
- Run:
code --install-extension opencode-zen-*.vsix
From Source
git clone https://github.com/SantiagoRoChi/OpenCode-Copilot-Chat.git
cd OpenCode-Copilot-Chat
npm install
npm run esbuild
npm run package
code --install-extension opencode-zen-*.vsix
🚀 Uso
Configuración Inicial
- Obtén una API key en opencode.ai/auth
- OpenCode local detectado automáticamente — usa tus keys existentes
- Configuración manual:
Ctrl+Shift+P → OpenCode Zen: Configure Zen Key
En Copilot Chat
- Abre Copilot Chat
- Selecciona un modelo del selector
- Los modelos aparecen agrupados por provider con context size y capabilities
Servidores Locales
# Iniciar servidor local
opencode serve --port 4096
# O añadir manualmente
Ctrl+Shift+P → OpenCode Zen: Add Server
Los servidores aparecen como proveedores separados en el selector de modelos.
Comandos
| Comando |
Descripción |
Configure Zen Key |
Configurar/limpiar Zen API key |
Configure Go Key |
Configurar/limpiar Go API key |
Add Server |
Añadir servidor OpenCode |
Edit Server |
Editar servidor existente |
Remove Server |
Eliminar servidor |
Refresh All Models |
Refrescar catálogo completo |
Show Output |
Ver logs de debug |
Clear Usage Stats |
Limpiar estadísticas |
🏗️ Architecture
extension.ts
├── providers/
│ ├── BaseProvider.ts # Base class (model caching, events)
│ ├── sdk/
│ │ ├── anthropicChat.ts # Anthropic SDK handler
│ │ ├── openaiChat.ts # OpenAI SDK handler
│ │ └── utils.ts # Shared message conversion
│ ├── LMStudioProvider.ts # LM Studio (OpenAI-compatible)
│ ├── OllamaProvider.ts # Ollama (OpenAI-compatible)
│ ├── OpenCodeFreeProvider.ts # Free tier models
│ ├── OpenCodeGoProvider.ts # Go tier models (Anthropic SDK)
│ ├── OpenCodeZenProvider.ts # Zen tier models
│ └── OpenCodeServerProvider.ts # Custom OpenCode servers
├── client/
│ ├── multiServerManager.ts # Server connection management
│ ├── modelRegistry.ts # Model capabilities from models.dev
│ └── types.ts
├── config/
│ └── secretStorage.ts # API key storage
├── integration/
│ └── opencodeConnector.ts # Auto-detect local OpenCode
├── tools/
│ └── subagentTool.ts # opencode_subagent tool
└── treeview/
└── openCodeTreeProvider.ts # Sidebar panel
Provider Architecture:
- All providers extend
BaseProvider (handles model caching, events, token counting)
- SDK-based handlers (
anthropicChat.ts, openaiChat.ts) use official AI SDK packages
- Shared utilities in
sdk/utils.ts eliminate code duplication
- LM Studio and Ollama use OpenAI-compatible API endpoints
📝 Recent Changes (2026-06-19)
AI SDK v6 Migration
- Architecture refactor: Migrated from custom HTTP streaming to official AI SDK
@ai-sdk/openai for OpenAI-compatible providers@ai-sdk/anthropic for Anthropic API providers
- Renamed:
OpenAICompatibleProvider → BaseProvider (reflects multi-provider support)
- Added: Shared utilities in
src/providers/sdk/utils.ts
convertMessages() - VS Code to AI SDK message formatmapModelOptions() - Temperature/topP/maxTokens mappingtrackToolNames() - Tool name tracking by callId
Code Cleanup
- Removed (~1000 lines of dead code):
src/providers/sdk/compatChat.ts - replaced by SDK handlerssrc/tools/toolCallAdapter.ts - replaced by AI SDK native handlingsrc/client/opencodeClient.ts - unused- Duplicate message conversion logic
- Cleaned: All
.js and .js.map files from src/ (build only in out/)
Bug Fixes
- Fixed tool schema format (use
jsonSchema() wrapper)
- Fixed tool call property (
input instead of args)
- Fixed reasoning blocks (
reasoningText instead of thinkingText)
- Fixed auth header caching (pass API key directly to SDK)
🚀 Nuevas Funcionalidades (v3.6.0)
📊 OpenCode Usage Tracking
- Datos de uso en tiempo real desde la API de OpenCode
- Burn-rate de suscripción Go: Seguimiento de 5h rolling, semanal y mensual
- Costo por modelo: Cálculo automático basado en tokens y precios
- Status bar mejorado: Muestra burn-rate con indicadores de advertencia
🤖 Agent Window Support
- Proveedores para Agents Window: Registra instancias duplicadas con sufijo -agent`r
- Configurable: \opencode-zen.enableAgentWindow\ (default: true)
- Copilot CLI: Modelos disponibles en Agents Window
🔐 OpenCode Auth Service
- Gestión centralizada: Workspace ID, cookie de auth, API keys
- Descubrimiento automático: Extrae workspace ID de páginas OpenCode
- Server ID dinámico: Descubre server ID del HTML de la página
🌐 OpenCode Usage Panel
- Simple Browser integrado: Login OAuth en panel de VS Code
- Servidor HTTP local: Captura workspace ID y cookie automáticamente
- Sin inputs manuales: Todo el proceso es automático
📈 Status Bar Mejorado
- Burn-rate de Go: Muestra 5h/weekly/monthly
- Indicadores de advertencia: ⚠️ cuando se acerca a límites
- Costo por request: Muestra costo estimado
🌳 Tree View Mejorado
- Comandos de registro: Login, Configure Workspace
- Datos de burn-rate: Muestra en dashboard
- Acceso directo: A todas las opciones de configuración
⚙️ Configuración
| Propiedad |
Default |
Descripción |
opencode-zen.requestTimeout |
60000 |
Timeout request (ms) |
opencode-zen.enableToolCalling |
true |
Tool calling |
opencode-zen.enableImageInput |
true |
Vision/image input |
opencode-zen.agentTemperature |
0.0 |
Temperatura para tools |
opencode-zen.verboseLogging |
false |
Logs detallados |
📊 Modelos Disponibles
OpenCode Zen (66 modelos)
GPT 5.x, GPT 5 Nano, Claude Opus 4.1-4.8, Claude Sonnet 4-4.6, Claude Haiku 4.5, Gemini 3.5 Flash, Gemini 3.1 Pro, Kimi K2.5/K2.6, DeepSeek V4 Flash, GLM 5/5.1, MiniMax M2.5/M2.7, Qwen 3.5/3.6/3.7, Grok Build 0.1, Big Pickle
OpenCode Go (16 modelos)
Kimi K2.5/K2.6, DeepSeek V4 Pro/Flash, GLM 5/5.1, MiMo V2.5/Pro, MiniMax M2.5/M2.7/M3, Qwen 3.5/3.6/3.7
OpenCode Free
DeepSeek V4 Flash Free, MiMo V2.5 Free, Nemotron 3 Super Free, Big Pickle
🔄 Auto-detección
- Detecta OpenCode local en
~/.local/share/opencode/auth.json
- FileSystemWatcher detecta nuevas keys
- Servidores OpenCode en ejecución se conectan automáticamente
📝 License
MIT
Hecho con ❤️ para desarrolladores que usan OpenCode
