Integrations

Wired into your stack.

Connect a model, point at your data, ship to a surface. Every call an agent makes through these is logged with its arguments and result, no exceptions.

Models

Anthropic

stable

Run Claude models as planners or verifiers. Tool calls and citations are logged line by line.

OpenAI

stable

Route reasoning to GPT-class models with per-step token and cost accounting captured in the trace.

Observability

Datadog

stable

Stream every trace as spans. Latency, cost, and verification rates land in your existing dashboards.

Surfaces

Linear

stable

Let agents file and triage issues. Each write-action requires a passed verification step.

Slack

stable

Ship agent output to channels with the trace attached, so reviewers can audit before acting.

Stripe

beta

Read billing context for support and revenue agents. Writes are gated behind human approval.

Data

Postgres

stable

Read-replica access with row-level guards. Agents query; TRACE records the exact SQL and result.

Snowflake

stable

Grounded answers from your warehouse. Every query an agent issues is recorded with its row count.

Extensible

Don't see it? Define it in minutes.

A tool is a typed function with a permission scope. Register it once and every agent can call it, with every call traced automatically.

Read the tools guide ▸

tools/refund.ts

defineTool({
  name: "issue_refund",
  scope: ["billing:write"],
  input: z.object({ id: z.string(), amount: z.number() }),
  run: async (i) => stripe.refunds.create(i),
});