What is Darkhunt?
Darkhunt supervises AI agents in production. It checks every step an agent takes, before it runs and after: the prompt, each tool call, what comes back and the response. Each step is weighed against the task, the conversation so far, the data involved and the agent’s normal behavior. Steps that fit go ahead. Your rules decide what happens to a step that doesn’t, and every decision is recorded.
What happens when a step doesn’t fit the task?
That depends on your rules. Darkhunt can flag the step and let it go ahead, stop it automatically or hold it for a person to review. A held step doesn’t run until they approve it. In every case, the agent keeps its access and the decision is recorded.
What does Darkhunt record?
Darkhunt records every step and what it decided about each one: what the agent did, the data and tools it touched and why the step was allowed, flagged or stopped. When a customer, your security team or a regulator asks how you keep agents under control, you send them the record.
How does Darkhunt connect to my agents?
Connect Darkhunt by SDK, gateway or plugin. It works with coding agents such as Claude Code, Codex and OpenClaw. It also works with Anthropic and Gemini models and with frameworks such as LangChain/LangGraph, the OpenAI Agents SDK and the Vercel AI SDK. Setup steps are in the setup docs ↗.
How is this different from a prompt firewall or guardrail?
Firewalls, guardrails and access controls judge one prompt, message or action at a time. Darkhunt checks each step against the task, the conversation so far, the agent’s normal behavior and your policy. An action can be permitted on its own and still be wrong for what the agent was asked to do.
Can Darkhunt test an agent before it goes live?
Yes. Before launch, Darkhunt can red team an agent with multi-turn attacks, such as instructions hidden in text or poisoned context, and follow each attack through the agent’s tool calls. Red teaming is included in every plan.
Can I run Darkhunt in my own environment?
Yes. Besides the hosted platform, Darkhunt can run inside your own infrastructure, so your agent’s traffic and the record stay in your environment. Regulated teams in healthcare and financial services, and anyone with data-residency rules, usually need this to get security sign-off.
Which frameworks and regulations does Darkhunt map to?
Each decision in the record is tagged to the frameworks you report against: HIPAA, OWASP Top 10 for LLMs, NIST AI RMF, MITRE ATLAS and ISO 42001. So you can answer a questionnaire or audit request by framework. EU AI Act mapping is in progress, planned for 2027. See the record.
Has Darkhunt been through a SOC 2 examination?
Yes. Darkhunt Limited completed a SOC 2 Type I examination (Security, Availability, Confidentiality) in August 2026. The Type II examination is in progress. Security documentation is in the Trust Center ↗, and you can request the full report through it.
How much does Darkhunt cost?
You pay in credits for what you use. Start on the Free plan, with credits included and no card needed. For an annual plan, we size the price to the agents you run and give you a real number. See plans and pricing.