LLM Endpoint Switchboard (by AIdome)

Configure your AI coding assistants to route through enterprise-approved LLM endpoints — in one click.

What is this?

The LLM Endpoint Switchboard is a configuration tool that helps enterprise teams redirect AI coding assistants to use approved LLM gateway endpoints. Instead of manually configuring each AI assistant separately, this extension manages endpoint configuration across all your AI tools from one place.

⚠️ Important: This Extension is NOT a Gateway

This extension does not proxy, relay, or intercept any API calls. It configures other extensions to point to YOUR gateway.

The value: One tool manages endpoint configuration across all your AI assistants — Continue, Cline, Roo Code, Kilo Code, Codex CLI, and more.

How It Works

┌─────────────────────────────────────────────────────┐
│                    VS Code                          │
│                                                     │
│  ┌──────────┐  ┌──────────┐  ┌──────────┐         │
│  │  Cline   │  │ Continue │  │ Roo Code │  ...     │
│  └────┬─────┘  └────┬─────┘  └────┬─────┘         │
│       │              │              │               │
│       └──────────────┼──────────────┘               │
│                      │                              │
│            ┌─────────▼──────────┐                   │
│            │    Switchboard     │                   │
│            │  (this extension)  │                   │
│            │  Configures each   │                   │
│            │  assistant's       │                   │
│            │  base_url + auth   │                   │
│            └─────────┬──────────┘                   │
│                      │                              │
└──────────────────────┼──────────────────────────────┘
                       │
              ┌────────▼────────┐
              │  Your Gateway   │
              │  (AIdome /      │
              │   LiteLLM /     │
              │   custom)       │
              └────────┬────────┘
                       │
          ┌────────────┼────────────┐
          │            │            │
     ┌────▼───┐  ┌────▼───┐  ┌────▼───┐
     │ OpenAI │  │Claude  │  │Gemini  │
     └────────┘  └────────┘  └────────┘

Supported Assistants

Claude Code gateway routing:

• Configures Claude Code user settings (~/.claude/settings.json by default, or CLAUDE_CONFIG_DIR/settings.json when overridden) with env.ANTHROPIC_BASE_URL for Anthropic Messages-compatible gateways.
• Requires raw OpenAI /v1/chat/completions endpoints to be fronted by a gateway that exposes Anthropic Messages, Bedrock InvokeModel, or Vertex rawPredict semantics.
• Enables CLAUDE_CODE_ENABLE_GATEWAY_MODEL_DISCOVERY=1; Claude Code uses that only on v2.1.129+ with Anthropic Messages gateways and only surfaces gateway models whose IDs begin with claude or anthropic.
• Keeps authentication outside plaintext config; supply credentials via ANTHROPIC_AUTH_TOKEN, ANTHROPIC_API_KEY, or Claude Code apiKeyHelper.

Tier Explanation

• Tier A — Full Automation: Extension automatically writes all configuration. One click.
• Tier B — Partial / Guided: Extension auto-discovers settings where possible, provides step-by-step guidance otherwise.
• Tier C — Informational: Extension detects the assistant, explains routing limitations, and suggests alternatives.

Dialects

Different AI providers use different API protocols. The Switchboard understands these and maps them correctly:

• openai.chat_completions — Most common, /v1/chat/completions (Continue, Cline, Roo Code, Kilo Code, CodeGPT, AnythingLLM)
• openai.responses — Newer OpenAI format, /v1/responses (OpenAI Codex CLI)
• anthropic.messages — Anthropic's /v1/messages (Claude Code)
• google.gemini.generate_content — Google Gemini (Gemini CLI)
• github.copilot — Routable via proxy override (debug.overrideProxyUrl). ⚠️ This setting is undocumented and may change without notice.
• tabnine.proprietary — Proprietary (not switchable)

Quick Start

1. Install the Extension

Install from the VS Code Marketplace or download the .vsix from GitHub Releases.

2. Run the Setup Wizard

1. Open Command Palette: Ctrl+Shift+P (Windows/Linux) or Cmd+Shift+P (macOS)
2. Run: AIdome: Setup Switchboard
3. Follow the wizard:
   • Detect installed assistants
   • Create an endpoint profile
   • Apply configuration
   • Verify connectivity ✅

3. Start Coding

Your AI assistants now route through your approved endpoint. No additional steps needed!

Commands

All commands available via Command Palette (Ctrl+Shift+P / Cmd+Shift+P):

Enterprise Safety

This extension is designed for enterprise environments with strict security requirements:

• 🔒 SecretStorage: Auth tokens stored via VS Code's encrypted SecretStorage API — never in settings files
• 📊 No Telemetry: Zero data collection. All processing is local.
• 💾 Backup Before Modify: Timestamped backups created before any configuration change
• 📋 Audit Trail: Full change log with undo capability. Export redacted diagnostics for support.
• 🔐 Secret Redaction: Diagnostics reports automatically redact all auth tokens, API keys, and sensitive data
• 🌐 Remote Aware: Detects SSH, Dev Containers, Codespaces, WSL — warns about reachability issues
• ↩️ Full Undo: Every change can be reversed. Factory reset available with confirmation.

Configuration

Profiles

Profiles store your endpoint configuration:

• Name: Human-readable identifier (e.g., "Production", "Staging")
• Base URL: Your LLM gateway endpoint (e.g., https://gateway.yourorg.com)
• Dialect: API protocol type (e.g., openai.chat_completions)
• Auth: API key (stored in SecretStorage, never in plain text)
• Tenant (optional): Organization or team identifier

Profiles are stored in VS Code's globalState and persist across sessions.

Managing Profiles

Use AIdome: Manage Profiles to:

• Create new profiles
• Edit existing profiles
• Delete profiles
• Switch between profiles

Advanced Runtime Settings

For slower enterprise networks, proxies, or remote hosts, the extension exposes advanced tuning knobs under the VS Code settings prefix aidome-switchboard.advanced.*.

• Tune CLI detection, HTTP timeouts, retry backoff, cache TTL, log buffering, and verifier timeouts from the Settings UI or settings.json.
• Environment variables override the matching setting when present, which is useful for managed shells and automation. Common overrides include HTTP_TIMEOUT_MS, AIDOME_SWITCHBOARD_CLI_DETECTION_TIMEOUT_MS, AIDOME_SWITCHBOARD_HTTP_RETRY_BACKOFF_MAX_MS, and AIDOME_SWITCHBOARD_VERIFIER_*.
• Keep the defaults unless you are troubleshooting proxy latency, certificate inspection, or remote reachability.

TLS Certificate Verification

By default the extension verifies TLS certificates on every HTTPS request (including the 7-step verification pipeline). For enterprise gateways that use self-signed or internal CA certificates, you can disable strict verification:

// settings.json
"aidome-switchboard.advanced.tlsVerify": false

Or via environment variable (useful in CI/automation):

export AIDOME_SWITCHBOARD_TLS_VERIFY=false

⚠️ Security Warning: Disable TLS verification only for trusted internal endpoints. The default is true and should remain enabled for production and public-internet endpoints.

When verification is disabled, the verifier TLS step reports a warning instead of pass/fail so administrators can audit which endpoints skip certificate checks.

Assistant TLS Verification Support

Each AI assistant handles TLS verification differently. The table below summarises how to control TLS verification for each supported assistant (the full data is in assistants.registry.json):

Generic Scanner

The extension includes a generic settings scanner for discovering unknown or future AI extensions:

• Uses heuristics to find settings keys in any extension
• Confidence scoring: High / Medium / Low
• Blocklist filtering to avoid false positives

Run AIdome: Setup Switchboard → Select "Scan for other AI extensions" to try it.

FAQ

Does this replace my AI assistant?

No. This is a configuration tool. You still need your AI assistants (Cline, Continue, etc.) installed.

Does this send my code anywhere?

No. This extension is purely a configuration manager. It doesn't proxy, relay, or see any of your code or API traffic.

What if my gateway goes down?

Your assistants will show connection errors. Use AIdome: Verify Routing to diagnose the issue.

Can I use this without AIdome?

Yes! It works with any OpenAI-compatible gateway: LiteLLM, custom proxies, or any /v1/chat/completions endpoint.

Does it work in remote environments (SSH, Containers, Codespaces)?

Yes. The extension detects remote environments and provides warnings if you configure a localhost endpoint that won't be reachable from the remote.

What about GitHub Copilot?

Copilot is supported at Tier B using the proxy override mechanism:

• Proxy Override (github.copilot.advanced.debug.overrideProxyUrl): routes all Copilot REST traffic (inline completions + chat) through your gateway.

⚠️ Note: debug.overrideProxyUrl is an undocumented internal setting and may change or be removed in future Copilot updates. Verify after major Copilot extension updates.

This setting is applied automatically and is fully reversible.

Not a Gateway — What This Extension IS and IS NOT

🚫 This Extension is NOT:

• ❌ An LLM provider or model host
• ❌ A proxy, gateway, or API relay
• ❌ A replacement for any AI assistant

✅ This Extension IS:

• ✅ A configuration manager for your existing AI tools
• ✅ A way to point all your assistants to one approved endpoint
• ✅ Enterprise-safe with encrypted secrets and full undo

Contributing

Contributions are welcome! Here's how to get started:

Development Setup

# Clone the repository
git clone https://github.com/AIdome-co/aidome-endpoint-switchboard-vscode.git
cd aidome-endpoint-switchboard-vscode

# Install dependencies
npm install

# Compile TypeScript
npm run compile

# Run tests
npm test

# Debug in VS Code
# Press F5 to launch Extension Development Host

Testing

npm test          # Run all tests
npm run lint      # Check code style
npm run compile   # Build TypeScript

Pull Request Guidelines

• Fork the repository and create a feature branch
• Write tests for new features
• Ensure all tests pass and code compiles
• Follow existing code style and conventions
• Submit PR with a clear description of changes

Report issues at: GitHub Issues

License

MIT License — see LICENSE file for details.

Copyright (c) 2026 AIdome

Note: This extension is a configuration tool for AI coding assistants. It is not affiliated with any specific AI assistant or LLM provider.