Captar gives OpenAI apps a runtime control layer that feels calm, precise, and easy to inspect.

Use the TypeScript SDK to reserve budget before a request, wrap your existing client instead of replacing it, track tools inside the same trace, and promote the best runs into datasets and manual eval review.

What ships in v1

Captar v1 focuses on traces, datasets, and manual review. The docs intentionally stay narrow so the SDK path is easy to follow and the platform story stays grounded in what is already available.

Start the quickstart

Create a session, wrap OpenAI, and record a tool call in one flow.

Learn the runtime model

See how budgets, policy, trace context, and usage reconciliation fit together.

See the operational loop

Inspect traces, export strong runs, and review them manually.

What Captar helps with

Keep spend local to the request instead of depending on a proxy or gateway.
Apply call and tool guardrails before work reaches a provider or external system.
Keep the span tree rich enough that trace review is useful later.
Move the best examples into datasets so the review loop stays repeatable.

Reading order

Overview