FAQ

Questions, answered.

Common questions about what Strathon does, how enforcement works, and how to run it. If something is missing, ask in the community or read the quickstart.

About Strathon

What is Strathon?

Strathon is an open-source firewall for AI agents. You write a policy as a CEL rule, and Strathon evaluates every tool call against it and blocks the dangerous ones before they execute. It runs in-process, in under a millisecond, and works across 10 agent frameworks.

How does Strathon block a tool call?

Strathon evaluates each tool call against your CEL policies at the tool-call boundary, before the underlying function runs. When a call matches a blocking policy, Strathon raises an error instead of executing it and records the decision in the audit log with the matched policy, trace context, and timestamp. Beyond blocking, it can also steer, throttle, redact, log, or require human approval.

Is Strathon a proxy?

The primary enforcement path is not a proxy. The SDK evaluates policies inside your agent process, so the core path adds no network hop. Strathon does ship an optional egress proxy as one of three enforcement layers, for governing raw outbound HTTP that an in-process SDK cannot see, but you opt into it; it is not required for the SDK or MCP gateway to work.

What is a CEL policy?

A CEL policy is a rule written in the Common Expression Language, the same expression language Google uses across its own infrastructure. Instead of static YAML lists, you express conditions over the tool call (its name, arguments, and context) and the action to take when they match. Strathon ships 12 templates mapped to the OWASP Agentic Top 10 so you can start from a working policy.

Using Strathon

Which agent frameworks does Strathon support?

Strathon supports 10 agent frameworks out of the box, including LangGraph, CrewAI, the OpenAI Agents SDK, LangChain, AutoGen, Pydantic AI, Google ADK, and the Claude Agent SDK. Integration is three lines of code: wrap your client, and the framework's tool calls pass the policy engine before executing.

How do I get started?

Start the server with Docker Compose, write a policy (or use a template), connect your agent with three lines of code, and watch a call get blocked. The quickstart walks through the full path from zero to a blocked tool call, and the community on Discord is there for questions.

How do I self-host Strathon?

Clone the repository, run Docker Compose, and open the dashboard. Strathon needs a single PostgreSQL database and runs anywhere you can run Docker; a Kubernetes pod spec and a from-source Python setup are also documented. The self-hosted firewall never phones home: no telemetry and no license check.

What does Strathon add in latency?

The in-process SDK adds sub-millisecond policy overhead because the evaluation happens inside your agent rather than over a network call. Actual throughput depends on your hardware and database, so Strathon ships a reproducible benchmark you run yourself instead of quoting a single headline number.

Security & compliance

What security threats does Strathon address?

Strathon is anchored on the OWASP Top 10 for Agentic Applications. It provides boundary enforcement against tool misuse, unexpected code execution, data and credential exfiltration, and insecure inter-agent communication, plus detection signals for memory poisoning, cascading failures, and behavioral drift. The honest scope, including what is enforceable at the tool-call boundary and what is detection-only, is documented rather than hidden.

Does Strathon detect leaked credentials and PII?

Yes. Strathon ships more than 70 credential patterns, detecting things like AWS keys, GitHub tokens, Stripe keys, and private keys, and redacts them at ingest. It also detects common PII entities so sensitive values do not end up in stored traces or get exfiltrated through tool calls.

Does Strathon help with EU AI Act compliance?

Strathon provides evidence export aligned to EU AI Act Articles 9 to 15 and 19, covering risk management, data governance, transparency, and human oversight, plus NIST AI Risk Management Framework risk scoring. It maps controls to these frameworks rather than certifying compliance, so you can demonstrate evidence to auditors without building bespoke tooling.

Pricing & cloud

Is the self-hosted version really free?

Yes, and there is no catch. The entire firewall (receiver, SDK, dashboard, every scanner and framework integration) is open source under Apache-2.0. Self-host as much as you want, with no request limits and no license keys.

When does managed cloud launch?

Q3 2026. We will announce a firm date when it is a few weeks away rather than commit to one we might miss. Until then, the open-source self-host path is the way to run Strathon, and it stays free forever. Join the Discord to be notified when cloud opens.

How much will managed cloud cost?

Pricing is not set yet and will be announced at launch. We would rather leave it blank than quote a number we cannot stand behind. Whatever it is, it will only cover hosting and support; the software itself is free.

Can I move from cloud to self-hosted later?

Anytime. It is the same software: export your policies and point your SDK at your own receiver.