Copilot Amplify

Extend GitHub Copilot Chat with the AI providers you choose.

Xiaomi MiMo · Z.ai GLM · Groq · NVIDIA NIM · OmniRoute — all first-class, equal siblings inside the VS Code Language Model API.

Install · Quick Start · Configuration · Contributing

✨ Why Copilot Amplify?

GitHub Copilot Chat is great — but you're limited to GitHub's model catalog. Copilot Amplify opens it up:

OmniRoute is a provider, not a middleman. Its internal multi-provider routing lives entirely on the OmniRoute server; this extension only ever talks to your configured base URL — exactly like any other vendor endpoint.

🗺️ Contents

• Installation
• Quick Start
• Supported Providers
• Architecture
• Configuration
• Commands
• Reliability
• Security
• Troubleshooting
• Development
• Contributing
• Documentation
• License

📦 Installation

Option A — build from source

git clone https://github.com/mojahid2021/copilot-amplify.git
cd copilot-amplify
npm install
npm run compile

Then press F5 in VS Code to launch an Extension Development Host.

Option B — packaged VSIX

npm install && npm run package   # → copilot-amplify-<version>.vsix

Install via Extensions: Install from VSIX….

Requires VS Code ≥ 1.125 and an active GitHub Copilot subscription (the extension extends Copilot Chat's Language Model API).

🚀 Quick Start

1. Open the panel — click the Copilot Amplify icon in the activity bar.
2. Add a key — click the 🔑 icon on any provider row and paste your API key. It goes straight into your OS keychain.
3. Pick a model — open Copilot Chat's model picker and choose e.g. Z.ai · Glm 5.2.
4. Chat. That's it.

No settings.json edits required. Full walkthrough with copy-paste recipes: docs/usage-examples.md

🔌 Supported Providers

Capabilities shown in the picker are truthful by design — a model is only advertised with vision/tools/reasoning if it actually supports them.

🏗️ Architecture

VS Code
    │
    ▼
VS Code Language Model API
    │
    ▼
┌─────────────────────────────────────────┐
│          Copilot Amplify Core           │
│                                         │
│  Provider Registry   Model Catalogs     │
│  Authentication      HTTP Transport     │
│  Streaming / SSE     Retry / Timeout    │
│  Circuit Breaker     Error Taxonomy     │
│  Context / Converter Thinking / Tools   │
│  Logging / Redaction Diagnostics        │
└────────────────────┬────────────────────┘
                     │
              Provider Registry
                     │
   ┌────────┬────────┼────────┬───────────┐
   ▼        ▼        ▼        ▼           ▼
 Xiaomi    Z.ai     Groq     NVIDIA    OmniRoute
 Provider  Provider Provider Provider  Provider
   │         │        │         │           │
   ▼         ▼        ▼         ▼           ▼
 MiMo API  Z.ai API Groq API NVIDIA NIM OmniRoute Server
                                                 │
                                     ┌───────────┼───────────┐
                                     ▼           ▼           ▼
                                 Provider A  Provider B  Provider C

Adding a provider means adding one descriptor module + tests — no core changes. See docs/adding-a-provider.md for the full guide.

src/
├── core/
│   ├── api/            # GenericApiClient (OpenAI SDK), shared SSE parser
│   ├── auth/           # SecretStorage-backed BaseAuthManager
│   ├── context/        # LM↔wire converters, token estimation & truncation
│   ├── errors/         # Normalized error taxonomy + HTTP mapping
│   ├── logging/        # OutputChannel logger with secret redaction
│   ├── models/         # Static model catalogs per fixed-catalog provider
│   ├── provider/       # BaseChatProvider + ProviderRegistry
│   ├── resilience/     # TTLCache (coalescing), CircuitBreaker, saturation classifier
│   ├── retry/          # Backoff, jitter, Retry-After, cancellation
│   ├── thinking/       # <think>/<thought> stream extraction
│   ├── diagnostics.ts  # Unified credential-free report across providers
│   └── url.ts          # Base URL validation / normalization / joining
├── providers/
│   ├── xiaomi/  zai/  groq/  nvidia/   # Fixed-catalog descriptors
│   ├── omniroute/                      # Full dynamic provider module
│   └── index.ts                        # Registry assembly
├── commands/           # Command registrations
├── ui/                 # Providers & Models tree view
├── extension.ts        # Activation (lightweight, zero network I/O)
└── secretsMigration.ts # One-time plaintext→SecretStorage migration
tests/
├── mocks/vscode.ts     # Minimal vscode API mock for unit tests
├── helpers/sse.ts      # SSE stream builders
└── unit/               # 21 files, 220+ cases incl. cross-provider contract tests

⚙️ Configuration

API keys & SecretStorage

Set keys through the UI — never edit settings files:

• Tree view → 🔑 icon on a provider row, or
• Copilot Amplify: Manage Providers… → Set API Key

Keys are stored via vscode.SecretStorage (OS keychain-backed) and never appear in logs, diagnostics, or errors. Upgrading from 1.x? Plaintext settings entries are migrated automatically on first activation and then removed.

💡 OmniRoute anonymous mode: no key needed for local servers — they work out of the box.

General settings

OmniRoute settings

Base URLs are normalized — these are all equivalent, no //v1 accidents:

https://example.com        https://example.com/
https://example.com/v1     https://example.com/v1/

🎮 Commands

General

OmniRoute-specific

🛡️ Reliability

• Retries — exponential backoff with full jitter on 408/429/5xx and network errors; honors Retry-After on every transport path; never retries 400/401/403/404, cancellations, or open circuits.
• Circuit breaker — per-provider, settings-driven; open circuits reject instantly instead of burning backoff time; automatic half-open recovery probe.
• Timeouts — separate request/discovery budgets; nothing hangs indefinitely.
• Saturation awareness — 503/529 bodies are inspected to distinguish queue saturation from maintenance; errors tell you which and what it means.
• Live health — connected · not-configured · error · rate-limited · auth-failed, reflected in the tree in real time.
• Streaming — hardened SSE parser: fragmented chunks, multi-event frames, CRLF, comments, [DONE], malformed-frame tolerance, trailing flushes. Reasoning deltas stream as native thinking parts.
• Cancellation — every request wires VS Code tokens into abort signals; cancelled streams release resources immediately.

🔒 Security

• 🔑 Credentials only in SecretStorage; legacy plaintext settings migrated away automatically and never logged.
• 🙈 Central logger redacts Authorization headers, bearer tokens, and sensitive keys; diagnostics expose only a boolean configured flag.
• 🌐 Base URLs validated before any network call: http/https only, embedded credentials rejected (SSRF guard).
• 🧰 Tool-call arguments parsed with prototype-pollution-safe JSON.
• 📋 Telemetry logging uses a strict header allowlist — nothing arbitrary is dumped.

🩺 Troubleshooting

More walkthroughs: docs/usage-examples.md §8.

💻 Development

npm install
npm run compile     # tsc build → out/
npm run watch       # incremental build
npm run lint        # eslint (src + tests, typed)
npm run typecheck   # tsc --noEmit (strict + noUncheckedIndexedAccess)
npm test            # vitest — 223 tests, fully mocked, no real keys needed
npm run package     # vsce → .vsix

The test suite covers URL normalization, SSE robustness, retry semantics, circuit-breaker transitions, cache coalescing, error taxonomy, secret redaction, tool-call parsing, context-truncation integrity, OmniRoute discovery/filtering, connection testing, session handling, telemetry normalization, secrets migration, and a cross-provider contract suite.

🤝 Contributing

Contributions are welcome! Good first steps:

1. 📖 Read docs/adding-a-provider.md — adding a provider requires exactly one descriptor module, registration wiring, and tests. No core changes.
2. 🐛 Check open issues or file a bug with reproduction steps.
3. ✅ Before opening a PR: npm run typecheck && npm run lint && npm test must pass.
4. 📝 Follow Conventional Commits (feat:, fix:, docs: …).

Design decisions are documented as ADRs in docs/adr/.

📚 Documentation

📄 License

ISC © Md Mojahid