Skip to main content
Tara is the observability assistant built into the Tracelane dashboard. Ask it a question about your own data — “what did my last session cost?”, “which traces after 3pm had errors?” — in plain language, by typing or by voice, and it answers by reading through the same tenant-scoped routes the dashboard itself uses.

​
What it does

  • Answers only from your own data. Every number, trace id or model name in an answer comes from a live read against your tenant’s traces, sessions, cost breakdown, latency/error summary, or guardrail history — never from anything Tara “remembers.” When it cannot find something, it says so instead of guessing.
  • Cites its sources. Answers carry citation chips linking straight to the traces they were built from.
  • Is itself a traced call. Asking Tara a question sends a request through your own gateway on your own connected provider key, exactly like any other chat completion — so the question and answer are captured, and you can open the conversation’s own trace from the panel.

​
What it does not do

  • Read-only. Tara cannot change, delete, annotate, or configure anything. It has no tools that write.
  • Not a general-purpose chatbot. Its tool set is a fixed list of read routes over your trace, session, cost, latency and guardrail data — it does not answer questions unrelated to your Tracelane data, and it will not always have an answer to every question about your data either (a narrow enough or unusual-enough question may return “I could not find that in your traces”).
  • No server-side voice. Speech-to-text and text-to-speech both run in your browser using the Web Speech API. Tracelane does not receive audio; your browser’s speech service may process it.

​
Voice support

Tara’s voice. When you press Read aloud, Tara prefers a natural female English voice from the ones your browser and operating system provide (for example Aria or Jenny on Edge, Google UK or US English Female on Chrome, Samantha on macOS and iOS). Voices are supplied by your device, not by Tracelane, so on a machine with no such voice installed the browser’s default is used. The microphone button uses your browser’s built-in SpeechRecognition API, and “read aloud” uses its built-in speech synthesis. Both are currently available in Chrome and Edge. In a browser without support, the microphone button is hidden with a tooltip explaining why — typing always works everywhere.

​
Where it lives

Open it from the Ask Tara button in the dashboard’s top bar. It needs at least one connected LLM provider key (Settings → LLM providers) — without one, the panel tells you so and links you to add one, without making a request.