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.
GitHub Copilot Chat is great — but you're limited to GitHub's model catalog. Copilot Amplify opens it up:
| 🔌 Bring your own provider | Xiaomi MiMo, Z.ai GLM, Groq, NVIDIA NIM, and OmniRoute work out of the box |
| ⚖️ No privileged provider | Every provider is an equal sibling — no hidden routing, no forced hierarchy |
| 🔐 Secrets done right | API keys live in VS Code SecretStorage (OS keychain), never in settings.json or logs |
| 🧠 Full AI capability surface | Streaming, tool calling, reasoning/thinking output, vision input, system prompts |
| 🛡️ Production-grade reliability | Retries with Retry-After, per-provider circuit breakers, timeouts everywhere, saturation-aware errors |
| 🌊 Live model catalogs | OmniRoute discovers models from your server at runtime — new models appear without updating the extension |
| 👁️ Observable | Unified diagnostics, per-request telemetry logs, live health states in the tree view |
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.
- Installation
- Quick Start
- Supported Providers
- Architecture
- Configuration
- Commands
- Reliability
- Security
- Troubleshooting
- Development
- Contributing
- Documentation
- License
Option A — from a marketplace (recommended)
Available on both major registries:
| Registry | Install |
|---|---|
| VS Code Marketplace | Install for VS Code — click Install, or inside VS Code: Extensions → Search "Copilot Amplify" |
| Open VSX | Install for VSCodium / Gitpod / Cursor & friends — or CLI: open-vsx compatible editors via code --install-extension mojahid2021.copilot-amplify |
Or straight from the terminal:
# VS Code / Cursor / Windsurf
code --install-extension mojahid2021.copilot-amplifyOption B — build from source
git clone https://github.com/mojahid2021/copilot-amplify.git
cd copilot-amplify
npm install
npm run compileThen press F5 in VS Code to launch an Extension Development Host.
Option C — packaged VSIX
npm install && npm run package # → copilot-amplify-<version>.vsixInstall 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).
- Open the panel — click the Copilot Amplify icon in the activity bar.
- Add a key — click the 🔑 icon on any provider row and paste your API key. It goes straight into your OS keychain.
- Pick a model — open Copilot Chat's model picker and choose e.g. Z.ai · Glm 5.2.
- Chat. That's it.
No settings.json edits required. Full walkthrough with copy-paste recipes:
docs/usage-examples.md
| Provider | Catalog | Highlights |
|---|---|---|
| Xiaomi MiMo | Static (2 models) | Auto-routes token-plan (tp-) keys to the dedicated gateway; up to 1M-token context |
| Z.ai GLM | Static (10 models) | Reasoning-effort control on GLM-5.x; vision on glm-5v-turbo |
| Groq | Static (7 models) | Ultra-fast Llama & GPT-OSS serving |
| NVIDIA NIM | Static (17 models) | Broad open-model catalog with transparent ID mapping |
| OmniRoute | 🔄 Dynamic | Discovers models live from GET /models; custom chat endpoints supported; anonymous local operation |
Capabilities shown in the picker are truthful by design — a model is only advertised with vision/tools/reasoning if it actually supports them.
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.
Project layout
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
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.
| Setting | Default | Purpose |
|---|---|---|
copilot-amplify.enableReasoning |
true |
Global reasoning toggle for thinking-capable models |
copilot-amplify.customSystemPrompt |
(empty) | Injected into every request across providers |
copilot-amplify.requestTimeoutMs |
120000 |
Chat request timeout for fixed-catalog providers |
copilot-amplify.debugLogging |
false |
Debug-level logs in the output channel |
copilot-amplify.circuitBreaker.enabled |
true |
Fail-fast after repeated failures — every provider |
copilot-amplify.circuitBreaker.failureThreshold |
5 |
Consecutive failures before a circuit opens |
copilot-amplify.circuitBreaker.resetTimeoutSeconds |
30 |
Cooldown before a half-open probe |
| Setting | Default | Purpose |
|---|---|---|
copilot-amplify.omniroute.baseUrl |
http://localhost:20128/v1 |
Server base URL (validated + normalized) |
copilot-amplify.omniroute.chatEndpoint |
(empty) | Custom chat-completions route — absolute URL or path relative to Base URL |
copilot-amplify.omniroute.noCache |
false |
Bypass the server-side semantic cache |
copilot-amplify.omniroute.noMemory |
true |
Skip server-side memory/skills injection |
copilot-amplify.omniroute.compression |
(empty) | Compression override (off, default, engine:<id>) |
copilot-amplify.omniroute.sessionId |
(empty) | Explicit session tag for cost attribution/memory |
copilot-amplify.omniroute.progress |
false |
Opt into server progress events |
copilot-amplify.omniroute.modelCacheTtlSeconds |
300 |
Model discovery cache TTL |
copilot-amplify.omniroute.requestTimeoutMs |
120000 |
Chat request timeout |
copilot-amplify.omniroute.discoveryTimeoutMs |
8000 |
Discovery / connection-test timeout |
copilot-amplify.omniroute.warmupOnStartup |
false |
Discover models at activation instead of first use |
copilot-amplify.omniroute.logTelemetry |
true |
Log routing/cost telemetry per request |
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/
General
| Command | Description |
|---|---|
Copilot Amplify: Manage Providers… |
QuickPick hub: keys, tests, refresh, diagnostics |
Copilot Amplify: Refresh Providers & Models |
Invalidate caches + re-discover |
Copilot Amplify: Show Diagnostics |
Credential-free report for all providers |
Copilot Amplify: Set / Clear API Key |
Credential management (context-aware in the tree) |
Copilot Amplify: Test Connection |
Lightweight reachability + auth probe |
Copilot Amplify: Pin / Unpin / Select Model |
Favorites and active model |
OmniRoute-specific
| Command | Description |
|---|---|
OmniRoute: Configure… |
Guided: base URL → chat endpoint → optional key → test |
OmniRoute: Edit Chat Endpoint |
Quick custom-endpoint override (also in the tree ACTIONS panel) |
OmniRoute: Set / Remove API Key |
SecretStorage credential management |
OmniRoute: Test Connection |
Latency + chat-model count report |
OmniRoute: Refresh Models |
Force re-discovery of /models |
OmniRoute: Show Diagnostics |
Full configuration & health report |
OmniRoute: Reset Configuration |
Restore defaults + clear stored key |
Show Omniroute Telemetry |
Open the per-request telemetry log channel |
- Retries — exponential backoff with full jitter on
408/429/5xxand network errors; honorsRetry-Afteron every transport path; never retries400/401/403/404, 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 —
503/529bodies 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.
- 🔑 Credentials only in SecretStorage; legacy plaintext settings migrated away automatically and never logged.
- 🙈 Central logger redacts
Authorizationheaders, 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.
| Symptom | Fix |
|---|---|
| "Authentication failed (401)" | Re-set the key (tree 🔑). Keys are never deleted automatically. 403 usually means region/access restrictions. |
| OmniRoute shows no models | Run OmniRoute: Test Connection. Check diagnostics for the last discovery error; missing /v1 in the Base URL is the usual culprit. |
| "circuit open" errors | The breaker tripped after repeated failures. Wait out the cooldown or lower failureThreshold; disable entirely via circuitBreaker.enabled. |
| New server model not in picker | Refresh Providers & Models. Non-chat entries (embeddings/rerank/TTS) are filtered by design. |
| Requests feel stuck | Lower requestTimeoutMs / omniroute.requestTimeoutMs; enable debugLogging and check the output channel. |
More walkthroughs: docs/usage-examples.md §8.
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 → .vsixThe 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.
Releases go to both registries:
npm run package # build copilot-amplify-<version>.vsix
# VS Code Marketplace (publisher: mojahid2021)
npx vsce publish
# Open VSX (VSCodium, Gitpod, and other VS Code forks)
npx ovsx publish --pat <OPEN_VSX_PAT>Contributions are welcome! Good first steps:
- 📖 Read docs/adding-a-provider.md — adding a provider requires exactly one descriptor module, registration wiring, and tests. No core changes.
- 🐛 Check open issues or file a bug with reproduction steps.
- ✅ Before opening a PR:
npm run typecheck && npm run lint && npm testmust pass. - 📝 Follow Conventional Commits (
feat:,fix:,docs:…).
Design decisions are documented as ADRs in docs/adr/.
| Doc | Contents |
|---|---|
docs/usage-examples.md |
Copy-paste setup recipes, endpoint examples, tuning, troubleshooting |
docs/adding-a-provider.md |
Step-by-step guide to contributing a new provider |
docs/adr/ |
Architecture decision records |
CHANGELOG.md |
Release history |
⭐ Found this useful? Star the repo — it helps others discover it!