🍴 Built as a fork of opencode-copilot-chat — the original OpenCode BYOK bridge — extended with Volcengine Ark and Qianwen AI as extra plan-based providers. For OpenCode Go/Zen specifics, see the upstream project.
💡 The pitch
Copilot Chat is great, but its premium models cost $39/mo (Pro+) and the free tier is rate-limited. This extension plugs multiple model gateways into the Copilot Chat model picker. You keep the native Copilot UI, tool-calling, and Agent Mode — you just get a much wider model catalog, and you can pick the cheapest provider per task.
🔌 Providers
The extension registers each provider as a separate vendor in VS Code's Chat → Manage Language Models. Any provider can be enabled, disabled, or removed from the picker independently — your API keys are kept, so re-enabling restores everything.
| Provider |
What it is |
Cost model |
Endpoints |
| OpenCode Go |
OpenCode's subscription gateway for curated open models (DeepSeek V4, Kimi K3, GLM-5.2, Qwen3.8 Max, MiMo V2.5, MiniMax M3) |
$10/mo subscription (5h/$12 · weekly/$30 · monthly/$60) |
OpenAI + Anthropic compatible |
| OpenCode Zen |
OpenCode's free-tier + pay-as-you-go gateway (Claude, GPT-5.x, Gemini, Grok, DeepSeek, rotating free models) |
Free models + pay-per-token premium |
OpenAI + Anthropic compatible |
| Volcengine Ark |
Volcengine's coding-plan endpoint (Doubao Seed, GLM-5.3, MiniMax M3, DeepSeek V4, Kimi K2.7) |
Coding-plan subscription |
OpenAI-compatible (/chat/completions) |
| Qianwen AI |
Alibaba's token-plan MaaS for Qwen models (qwen-max, qwen-plus, qwen-turbo) |
Token-plan subscription |
Anthropic-compatible (/v1/messages) |
💸 Why plan-based providers? Per-token pricing on frontier models adds up fast during long agentic sessions. Volcengine Ark (coding plan) and Qianwen AI (token plan) let you use the same underlying models under a flat plan instead of per-token metering — wire them once and route heavy workloads there.
🧠 Models
Models are fetched live from each provider on startup (Qianwen AI exposes a live /models list; Volcengine Ark uses a static list because its coding plan has no /models endpoint), with a bundled offline fallback.
OpenCode Go (subscription)
DeepSeek V4 Pro/Flash, Kimi K3 / K2.7-code / K2.6, GLM-5.2 / 5.1, Qwen3.8/3.7/3.6/3.5 Max+, MiMo V2.5 / V2.5-pro, MiniMax M3 / M2.7, GPT-5.6 Luna, Hy3 — with per-family thinking controls and generous context up to ~1M tokens.
OpenCode Zen (free + pay-as-you-go)
Rotating free models (Big Pickle, DeepSeek V4 Flash Free, MiMo V2.5 Free, Hy3 Free, Nemotron 3 Ultra Free, …) plus paid Claude Opus/Sonnet/Haiku, GPT-5.x, Gemini 3.x, Grok 4.5, DeepSeek V4, Kimi K2.6, MiniMax M2.7, Qwen3.6+, and more.
Volcengine Ark (coding plan)
Default list (overridable via volcengineArk.models):
doubao-seed-evolving · doubao-seed-2.1-turbo · doubao-seed-2.0-lite · minimax-m3 · glm-5.3 · glm-5.3-flash · deepseek-v4-flash · deepseek-v4-pro · kimi-k2.7-code
Qianwen AI (token plan)
Live list with these built-in fallbacks (overridable via qianwenai.models):
qwen-max · qwen-plus · qwen-turbo · qwen-max-latest · qwen-plus-latest
Context & output limits resolve per model (live metadata → models.dev snapshot → bundled fallback). Deprecated/unavailable models are filtered from the picker automatically.
⚡ Quick Start
1. Install or update VS Code 1.125+ ───────────────────────────── ✓
2. Install this extension ──────────────────────────────────────── ✓
3. Get an API key from at least one provider (below) ───────────── ✓
4. Copilot Chat → model picker → "Add Models" → pick your provider ✓
5. Paste the key → select a model → CHAT 🎉
Get a key:
- OpenCode Go / Zen — sign up at opencode.ai. Zen starts free (rotating free models); Go is a $10/mo subscription.
- Volcengine Ark — Volcengine console → coding plan → create an API key.
- Qianwen AI — Alibaba Cloud Model Studio → token-plan MaaS → create an API key.
📖 Detailed step-by-step
- Install or update VS Code to 1.125+. BYOK chat works without a GitHub sign-in or Copilot plan.
- Install this extension from the Marketplace (or press
F5 in this repo for dev mode).
- Get an API key from one of the providers above.
- Open Copilot Chat (Cmd/Ctrl+Shift+I).
- Click the model picker → Add Models…
- Select OpenCode Go, OpenCode Zen, Volcengine Ark, or Qianwen AI.
- Paste your API key when prompted (stored by VS Code in its encrypted language-models storage — it never leaves your machine).
- Pick the models you want enabled.
- Select any model from the picker and start chatting. 🚀
💡 Tips:
- Providers are independent groups — add several and switch anytime from the picker.
- If a model shows in Language Models but not the chat picker, hover its row and click the eye icon (👁) to enable it.
- Set
opencodego.freeOnly: false to reveal paid OpenCode Zen models in the picker.
✨ Features
- Multi-provider routing — every model family auto-routes to its native transport (
/responses, /messages, /chat/completions, streamGenerateContent) with per-endpoint tool-calling formats.
- 🧠 Thinking controls — per-model reasoning effort (DeepSeek
off→max, Qwen thinking_budget, GLM/Kimi/MiniMax/MiMo toggles and levels).
- 🖼️ Vision proxy — text-only models can "see" images via a configured vision model; per-image descriptions are cached and reused across turns.
- 📊 Go usage tracking — status bar burn-rate across 5h / weekly / monthly tiers (server-synced from OpenCode Go's usage endpoint).
- 🪟 Agents window (Copilot CLI) — all providers appear in the Agents-window model picker via agent-host variants.
- 🖼️ Vision + PDF + Audio — multimodal models pass through image, PDF, audio, and video inputs (oversized images auto-resize to 2000×2000 / 5MB).
- 📐 Context-size picker — tiered-context models expose
256K vs full-window selection, cheaper tier default.
- 🔌 Provider on/off — remove or re-add any provider from Language Models & every picker; keys and BYOK groups are kept.
- ✍️ Inline suggestions (experimental) — opt-in ghost-text completions with thinking forced off.
- 🛠️ Reliability — transient 5xx retry with backoff, sticky gateway headers, request/stream timeouts, context-overflow auto-retry.
🔧 Settings
All settings live under the opencodego.*, volcengineArk.*, and qianwenai.* namespaces. Key ones:
| Setting |
Default |
Description |
opencodego.apiBaseUrl |
https://opencode.ai/zen/go/v1 |
Base URL for the Go-compatible gateway |
opencodezen.apiBaseUrl |
https://opencode.ai/zen/v1 |
Base URL for the Zen-compatible gateway |
volcengineArk.apiBaseUrl |
https://ark.cn-beijing.volces.com/api/coding/v3 |
Volcengine Ark coding-plan base URL (OpenAI-compatible) |
volcengineArk.models |
(built-in list) |
Comma-separated model-ID override |
qianwenai.apiBaseUrl |
https://token-plan.cn-beijing.maas.aliyuncs.com/apps/anthropic |
Qianwen Anthropic-compatible Messages API base URL |
qianwenai.modelsBaseUrl |
https://token-plan.cn-beijing.maas.aliyuncs.com/compatible-mode/v1 |
Qianwen live-model-list base URL |
qianwenai.models |
(built-in list) |
Comma-separated model-ID fallback |
opencodego.freeOnly |
true |
Zen: free models only. false = include paid |
opencodego.temperature |
0.2 |
Sampling temperature (0–2) |
opencodego.maxTokens / maxInputTokens |
0 |
Max output / context override (0 = per-model default) |
opencodego.stripThinkTags |
auto |
Strip thinking tags (never/auto/always) |
opencodego.agentsWindow |
true |
Expose agent-host model variants for the Agents window |
opencodego.thinking.* |
off |
Per-family reasoning effort defaults |
Run Preferences: Open Settings (UI) and search opencode / volcengine / qianwen for the full list.
🎛️ Commands
Run these from the Command Palette (Cmd/Ctrl+Shift+P). Most are registered per provider — OpenCode Go, OpenCode Zen, Volcengine Ark, and Qianwen AI each expose the same set:
| Group |
Commands |
| Per provider (×4) |
…: Manage Provider — test connection, refresh models, configure utility models
…: Refresh Models — force a fresh model-list fetch
…: Diagnostics — markdown report of models + recent request summaries
…: Remove/Re-add Provider in Language Models — toggle the provider off/on in every picker |
| Cross-provider |
OpenCode: Model Picker Diagnostics — all registered models side-by-side
OpenCode: Configure Utility Models — utility-task model settings
OpenCode: Set Thinking Effort… — per-family reasoning picker |
| OpenCode Go |
OpenCode Go: Show Usage Details · …: Show Usage Quick Pick · …: Set Usage Targets… · …: Rename Active Profile · …: Delete Profile — subscription metering & profiles
OpenCode Go: Configure Vision Proxy — pick a vision model so text-only models can "see" images |
❓ FAQ
Do I need Copilot Pro, Pro+, or Max?
No. BYOK chat works without a Copilot plan and without signing in to GitHub. Requests are billed only by the provider you configured and do not consume Copilot requests. (Inline suggestions, next-edit suggestions, and semantic search still require Copilot.)
Where is my API key stored?
In VS Code's language-models configuration (VS Code stores it in encrypted storage). It never leaves your machine and is only sent to the provider you configured.
Can I use multiple providers at the same time?
Yes. Each provider is a separate group. Add several via Language Models → Add Models…, enter each key separately, and switch between them from the picker anytime.
How do I use these models in the Agents window (Copilot CLI)?
Agent-host variants are enabled by default (opencodego.agentsWindow: true). Open the Agents window, start a Copilot CLI session, and pick any provider's model from the picker.
How do I report a bug or request a model?
Open an issue — pick the Bug Report or Feature Request template, and include the diagnostics report (OpenCode Go: Diagnostics / OpenCode Zen: Diagnostics).
🤝 Contributing
Contributions welcome — typo fixes, new model support, or screenshots. See CONTRIBUTING.md for guidelines and the development workflow.
Press F5 in VS Code to open an Extension Development Host:
npm install # install deps
npm run compile # build TypeScript
npm run watch # watch mode
npm run package # build .vsix
If this saves you money or unlocks a model you needed — ⭐ star the repo and share it!
📄 License
MIT © 2026 ltmoerdani, zphilip
OpenCode is a trademark of opencode.ai. This project is independent and not affiliated with GitHub, Microsoft, Anthropic, OpenAI, Google, Alibaba, Volcengine, or any model provider.
