Simple pricing built around your agent footprint.

An AI agent counts as one unit if it runs as an independent process with its own tool access context. Multiple instances of the same agent type each count separately. Chains where one agent calls another count each active agent in the chain. Sub-tasks within a single agent session that share the same context do not multiply the count. Your plan tier's agent cap applies to concurrent active agents, not total sessions over a period.

Yes. If your concurrent agent count exceeds your current plan tier, the Arrakis console will surface an overage notification. You can upgrade to the next tier at any time and the change takes effect immediately. Overage charges are prorated to the remaining billing period. There is no hard cutoff -- agents above your tier limit continue to be monitored, and you are billed for the upgrade at the end of the cycle.

Arrakis instruments LangChain, CrewAI, AutoGen, the OpenAI Assistants API, and custom agent frameworks built on raw LLM client calls (OpenAI, Anthropic, Cohere, and others). The Python SDK supports any framework that can accept a wrapper around its executor or tool-call handler. The HTTP proxy path works regardless of framework as long as the agent makes outbound HTTP calls. TypeScript SDK support is in early access for LangChain.js and custom Node.js agents.

A self-hosted deployment option is on the roadmap for Enterprise customers with strict data residency requirements. In the current release, Arrakis processes agent event metadata (action types, resource targets, policy outcomes, timestamps) in our managed cloud. We do not process the content of agent inputs or outputs unless you opt in for debugging purposes. Custom data residency (US or EU region selection) is available on Enterprise plans for metadata processing.

The 14-day free trial gives you full access to your selected plan tier with no restrictions. You can instrument real agents, define policies, and see live monitoring. No credit card is required for the Starter trial. For Professional trials, a card is required but will not be charged until the trial ends. Trials for Enterprise plans are arranged through our sales team and include a guided onboarding session.

Enterprise audit logs are retained for one year in an immutable append-only store. Each log entry contains: agent identifier, session ID, timestamp (UTC), action type, resource target, policy evaluated, policy outcome (permitted/blocked/throttled), and the triggering context (user-initiated vs. scheduled vs. inter-agent). Logs are exportable in structured JSON and CSV. For SOC 2 evidence packages, we provide a structured export scoped to a defined audit period. GDPR data export for subject access requests is also available on Enterprise plans.