feat: add Vercel AI SDK Sentry instrumentation and document OPENAI_AP…

[KyleTryon]

This pull request introduces optional AI-powered features to TuvixRSS, specifically AI-based category suggestions for RSS feeds using OpenAI and the Vercel AI SDK. It also adds comprehensive Sentry instrumentation for monitoring AI usage, latency, and errors. The changes include updates to documentation, environment configuration, and code to support secure and observable AI feature rollout.

AI Features & Configuration:

• Added AI-powered category suggestion feature using OpenAI's GPT-4o-mini model, gated by global settings, user plan, and API key presence (aiEnabled flag, Pro/Enterprise plan, OPENAI_API_KEY). Documentation and environment examples updated for local, Cloudflare, and CI/CD deployments. [1] [2] [3] [4]

Observability & Sentry Integration:

• Integrated Vercel AI SDK with Sentry for automatic tracking of AI calls, including token usage, latency, model info, and errors, with configuration for both Node.js and Cloudflare Workers environments. [1] [2] [3]
• Enabled detailed telemetry for AI calls in the category suggester service, capturing inputs, outputs, and function identifiers for better traceability.
• Updated Sentry initialization logs and configuration to indicate AI tracking is enabled. [1] [2]

[KyleTryon]

@sentry review

[Copilot]

Pull request overview

This PR adds Sentry instrumentation for AI-powered features using the Vercel AI SDK, enabling automatic tracking of OpenAI API calls for category suggestions. The changes include configuring the Sentry Vercel AI integration, enabling telemetry in AI SDK calls, and documenting the OPENAI_API_KEY environment variable for local, Cloudflare, and CI/CD deployments.

Changes:

• Integrated Sentry's vercelAIIntegration in Node.js and Cloudflare Workers entry points to track AI SDK calls
• Enabled experimental_telemetry in the AI category suggester service to capture token usage, latency, model info, and errors
• Added OPENAI_API_KEY documentation to env.example, deployment.md, and CLAUDE.md with setup instructions for different deployment methods

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
packages/api/src/services/ai-category-suggester.ts	Added experimental_telemetry configuration to AI SDK call for automatic span tracking
packages/api/src/entries/node.ts	Configured Sentry vercelAIIntegration with recordInputs/recordOutputs enabled
packages/api/src/entries/cloudflare.ts	Added vercelAIIntegration to Sentry config and enabled AI tracking in initialization logs
packages/api/src/config/sentry.ts	Added enableAIIntegration configuration field and comments about AI SDK integration
env.example	Documented OPENAI_API_KEY environment variable with setup instructions
docs/deployment.md	Added OPENAI_API_KEY configuration for Docker and Cloudflare deployments
CLAUDE.md	Added comprehensive AI Features Configuration section with Sentry instrumentation details

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

There's an inconsistency between Node.js and Cloudflare configurations: Node.js sets recordInputs: true, recordOutputs: true at the integration level, while Cloudflare doesn't pass these parameters to vercelAIIntegration(). Both rely on experimental_telemetry settings in individual AI SDK calls. For consistency and clarity, consider either: (1) removing the parameters from Node.js integration to match Cloudflare's approach and only control recording via experimental_telemetry, or (2) adding the same parameters to Cloudflare's integration. The current mixed approach may cause confusion about which settings take precedence.

Suggested change

	// Note: Input/output recording is controlled via experimental_telemetry in AI SDK calls
	config.integrations = [Sentry.vercelAIIntegration()];
	// Note: Input/output recording is enabled here but still controlled per call via experimental_telemetry in AI SDK usage
	config.integrations = [
	Sentry.vercelAIIntegration({
	recordInputs: true,
	recordOutputs: true,
	}),
	];