Skip to content

ai-providers.md

Latest commit

 

History

History
93 lines (62 loc) · 4.7 KB

ai-providers.md

File metadata and controls

93 lines (62 loc) · 4.7 KB

AI Providers

Team X-Ray analyzes local git data and then chooses the AI path that the current implementation supports. The main flow is Copilot SDK first, GitHub Models fallback second, with a reduced local fallback only if AI output cannot be produced.

Mode Setting value What it does Requirements
Copilot SDK copilot Default analysis path with custom tools over local repo data Copilot CLI installed + authenticated; set teamxray.cliPath if needed
BYOK via Copilot SDK byok-openai, byok-anthropic, byok-azure Applies a provider override to the Copilot SDK session Copilot CLI, Team X-Ray: Set BYOK API Key (Secure), teamxray.byokBaseUrl, teamxray.byokModel
GitHub Models fallback github-models Uses your GitHub token when Copilot is unavailable or analysis falls back Team X-Ray: Set GitHub Token
Reduced local fallback — Builds a basic git-derived analysis if AI output fails No extra setup

Copilot SDK

This is the primary path. Team X-Ray dynamically loads @github/copilot-sdk, creates a CopilotClient, registers custom tools with defineTool, and sends repository data into a single analysis session.

ESM bundling note: the SDK is ESM-only, so the dynamic import must target @github/copilot-sdk directly:

const sdk = await import(/* webpackIgnore: true */ '@github/copilot-sdk');

If the CLI is installed outside your PATH, point the extension at it with teamxray.cliPath.

BYOK provider overrides

BYOK (Bring Your Own Key) lets you point the Copilot SDK session at a different model provider using your own API key. Use it when:

  • Your org blocks Copilot but permits direct provider access
  • You want a specific frontier model (e.g. a newer Claude release)
  • You route AI traffic through an internal proxy or gateway
  • You're testing model-to-model quality on the same repo

Important: Copilot CLI is still required

BYOK is wired through the Copilot SDK session, not as a standalone provider path. The Copilot CLI must be installed and authenticated even when teamxray.aiProvider is set to a byok-* value. Your API key, baseUrl, and teamxray.byokModel override the model the SDK talks to; the session orchestration is still Copilot's.

If this changes in a future version, this note will be removed.

Supported providers

teamxray.aiProvider Provider Typical byokBaseUrl
byok-openai OpenAI https://api.openai.com/v1
byok-anthropic Anthropic https://api.anthropic.com/v1
byok-azure Azure OpenAI https://YOUR-RESOURCE.openai.azure.com/openai/deployments/YOUR-DEPLOYMENT

Azure URLs include the resource name and deployment name. Get them from the Azure Portal under your OpenAI resource → Deployments.

End-to-end example (Anthropic)

  1. Install and authenticate the Copilot CLI (see Option 1 above).
  2. Command Palette: Team X-Ray: Set BYOK API Key (Secure). Paste your sk-ant-... key.
  3. Open VS Code settings (JSON) and add: 
    "teamxray.aiProvider": "byok-anthropic",
"teamxray.byokBaseUrl": "https://api.anthropic.com/v1",
"teamxray.byokModel": "claude-sonnet-4-5-20250929"
  4. Run Team X-Ray: Analyze Repository Expertise.

Settings reference

Setting Required Notes
teamxray.aiProvider yes Set to byok-openai, byok-anthropic, or byok-azure
API key (SecretStorage) yes Set via Team X-Ray: Set BYOK API Key (Secure), not in JSON settings
teamxray.byokBaseUrl yes Full endpoint URL. See table above
teamxray.byokModel yes Model identifier. Examples: gpt-4o, claude-sonnet-4-5-20250929, your Azure deployment name

Migrating from teamxray.byokApiKey

The teamxray.byokApiKey setting is deprecated because settings JSON is plaintext. If you have a value there from an older version:

  1. Run Team X-Ray: Set BYOK API Key (Secure) and paste the same key. It will be stored in VS Code's encrypted SecretStorage.
  2. Remove teamxray.byokApiKey from your settings JSON.

The extension reads SecretStorage first and only falls back to the plaintext setting for migration.

GitHub Models fallback

If the Copilot SDK is unavailable, Team X-Ray falls back to GitHub Models using your GitHub token.

Store the token with:

Command Palette → Team X-Ray: Set GitHub Token

The current implementation calls https://models.github.ai/inference/chat/completions.

Reduced local fallback

Team X-Ray does not currently expose a normal selectable "local-only" tier for the main analysis flow. Instead, it can assemble a reduced result from local git history when AI analysis cannot complete or the AI response cannot be parsed cleanly.