Pricing

Priced per agent. Never per event.

An audit product billed per event pays you to instrument less. So recording everything costs the same as recording something: one price per agent, unlimited actions, and the rational move at every tier is complete coverage.

Self-serve

Free
while you evaluate

Install the SDK, instrument a real agent, verify a real bundle. This is where developers start.

  • All four capture planes
  • Hash chain and signed checkpoints
  • Offline verification, always
  • Native spend analysis and budgets
  • Community support
Start free
Most common

Team

$99
per agent / month · unlimited recorded actions

Every action your agent takes is chained, checked, and kept — recording more never costs more, because an audit product must never make you instrument less.

  • Everything in Self-serve
  • Unlimited events per agent — coverage is free at the margin
  • Policy enforcement and one-click human sign-off
  • Examiner packet: FINRA · EU Art 12 · Colorado · AIUC-1
  • Public trust page for your security reviews
  • Cost per agent, per run, per customer
  • Multi-year retention
  • External timestamping (RFC 3161)
Start with one agent

Enterprise

Contract
annual, volume-based

For regulated deployments that run it themselves and bring their own security review.

  • Everything in Team
  • Self-hosted — runs entirely in your VPC
  • Metadata-only mode (we never see payloads)
  • WORM storage and legal hold
  • SSO, audit log export, custom DPA
  • Named support with an SLA
Talk to us
Questions people actually ask

The ones that decide it.

Including the one most pricing pages avoid: what happens to your evidence when you stop paying for it.

Start free
01
Why per agent and not per event?
Because per-event billing on an audit product pays you to instrument less, and incomplete coverage is the one failure this category cannot survive. Per agent, the marginal cost of chaining everything is zero — which is the behaviour the record needs from you, priced accordingly.
02
What counts as an agent?
One deployed agent identity — the actor id on its events. A fleet of twelve support bots is twelve agents; the same bot in staging and production is two. Retries and internal bookkeeping are never billed, and if we cannot price an upstream call the record says so rather than guessing.
03
Do you see our prompts?
Not by default. The standard mode records metadata and cryptographic hashes only; payloads stay in your storage. Integrity is still fully provable, because the hashes are what chain.
04
What happens if we stop paying?
You keep your evidence. Export is always available and the bundles verify without us — that is the point of offline verification. A compliance record you lose access to is not a compliance record.
05
Can we run it ourselves?
Yes, on Enterprise. Nothing in the architecture assumes our cloud: the chain, the policy engine and the verifier all run inside your VPC. External anchors work from anywhere.
06
How is this different from our observability tool?
Those answer “is the model behaving” for engineers, on mutable logs with short retention. This answers “prove what the agent did and that the record is unaltered” for the person who has to sign.