Latten

What is AI observability for a company running agents and copilots?

AI observability is the live view of what every assistant, agent, and model call is doing: who triggered it, which model ran, what data or service it reached, and what it cost. For company AI, logs are not enough. You need cost and reach attributed to the actor behind each call, so adoption can keep moving without turning into an opaque bill or an unknown access path.

See it on a live graph — no signup →

Observe the boundary crossings

The important moments are where AI crosses a boundary: an LLM call, a database read, an external API, or an agent handoff. Latten records those crossings (each time your AI touches data, a model, or a tool) and renders them as one live authority graph.

Tie cost to the actor

A token total is only a bill. Latten attributes input, output, cached, and reasoning tokens to the actor, model, and data domain behind the call, so cost has an owner.

Tie access to real traffic

Architecture diagrams drift. Latten maps reach from the traffic your AI actually generates, read-only and out of the request path.

How it works

  1. 1. Seed the actor Identify the person, agent, or service account behind a request.
  2. 2. Track each crossing Wrap LLM calls, data touches, external APIs, and handoffs.
  3. 3. Read the graph See cost and reach together, per actor and per data domain.
  4. 4. Act on evidence Fix loops, orphaned sessions, and overbroad reach after they are visible.

Common questions

Is AI observability the same as LLM monitoring?

No. LLM monitoring usually focuses on prompts, latency, and token usage. Company AI observability also shows who acted and what company systems the AI reached.

Does Latten sit in the request path?

No. It is observe-only and reads traffic without changing how the app behaves.

Can this work with TypeScript AI apps?

Yes. Use @latten/sdk directly or the @latten/installer MCP to place the SDK in a reviewable pull request.