Skip to main content

​
Which runtimes are supported?

The Node SDK requires the Node.js runtime (Node 18+). The Edge runtime on Vercel and similar platforms won’t work because OpenTelemetry needs Node APIs. The Python SDK runs on CPython 3.10 or newer. See Vercel for the runtime opt-in.

​
Do I need to set baseUrl?

No. It defaults to https://platform.belvedir.ai. Only set it if you’re targeting a self-hosted or staging platform host.

​
How long until traces appear in the dashboard?

Traces appear within seconds. Tasks and training sets form ~30 seconds after a session goes quiet; segmentation only runs once a session is idle.

​
Is my API key safe in instrumentation.ts?

Yes, instrumentation.ts runs on the server. Never expose BELVEDIR_API_KEY to client-side code.

​
Can I use one project for multiple apps?

Yes. Set a different appName per app when calling initialize(); each app’s traces carry it in their stored metadata. To split apps apart in the dashboard’s views, give each app its own API key: the Data pages filter by key.

​
What if my agent uses more than one LLM provider?

Pass every provider you use to instrumentModules, e.g. { openAI: openai.OpenAI, anthropic }. Every span records which provider and model served it, so you can compare them in the dashboard.

​
Can I route between local and hosted models?

Yes, see Model Routing.

​
Can I tell Belvedir whether a session actually worked?

Yes, report ground-truth outcomes with reportOutcome(sessionId, "success" | "fail") (Python: report_outcome), or POST /api/v1/outcomes directly. A reported fail keeps that session’s tasks out of training data. Report after flush(): an outcome for a session the platform hasn’t ingested yet returns false/404. See the SDK API.

​
What does Belvedir charge for?

Only inference: routed calls through the router endpoints and hosted inference sessions, per token at the rates on the platform’s Pricing page (Inference → Pricing), drawn from the organization’s credit balance. Tracing, segmentation, training runs, and benchmark runs are currently free.

​
Why did I get a 402 or 429?

A 402 means the organization is out of credits (or the project has no organization); a 429 is the per-key rate limit. Both are covered in Common Issues, with the limits and the fixes.