The one thing here that is a product, not a service.
Everything else on this site is people doing work on your bill. Tidal Telemetry is software you sign up for and run yourself. It is free during beta, it requires no engagement with us, and this page exists so the difference is not confusing.
What it is
Hosted LLM observability. You install the SDK, call tidal.init(), and every model call your application makes starts arriving as a trace: model, token counts in and out, an estimated cost per call, latency, errors.
- One line, roughly 50 libraries. Auto-instrumentation covers OpenAI, Anthropic, Mistral, Bedrock, LangChain, LlamaIndex, vector databases and the common agent frameworks.
- Standard OpenTelemetry, so the SDK is optional. Ingest is plain OTLP over HTTPS with a bearer token. Point any OpenTelemetry exporter at your project endpoint and it behaves identically. The SDK is a thin wrapper over the Apache-licensed OpenLIT SDK.
- Isolated per-project storage. Each project gets its own database and its own ingest key, enforced at the storage layer rather than by a filter in a query.
- An MCP server. Claude and other agents can check your setup, send a test trace and pull usage reports, which is a faster debugging loop than reading a dashboard.
Python and Node get the one-line install. Anything that speaks OpenTelemetry can export directly.
What it costs
Nothing, currently. It is in beta and everything is free, with paid tiers planned and a free plan kept for beta users. That is the actual state of it rather than a launch offer with a countdown attached.
How it relates to the service
Telemetry shows you what your LLMs spend. The managed service is the engineering work that reduces it, priced on verified savings. They are cross-promoted and deliberately not bundled:
- You can use Telemetry and never talk to us. There is no sales gate on it and no engagement prerequisite.
- You can hire us and never use Telemetry. Engagements work against your provider invoices and whatever tooling you already run. Nothing in our pricing licenses a dashboard to you, and we do not price one in.
- Below the $20k/month engagement minimum, this is the better answer. Instrument it yourself, read your own numbers, and use the free templates to allocate them.
Keeping these separate is a constraint we put on ourselves for a reason: a cost consultant who also owns the meter has an obvious problem, so the baselines behind every fee we charge come from raw provider invoices and not from our own product.