Answers about Gensee Crate, including what it defends against, how sidecar deployment works, how it handles long-horizon risk, and how enterprise deployments keep policy and evidence on-prem.
The existing GenseeAI FAQ covers agent hosting. This page is specifically for Gensee Crate, the runtime security product.
Gensee Crate is runtime defense for systems that use AI agents. It observes user requests, agent behavior, tool calls, memory, skills, files, network activity, processes, and execution so unsafe agent behavior can be detected before it becomes a system-level side effect.
Gensee Crate defends against prompt injection, memory poisoning, long-horizon attacks, risky tool use, unsafe file or network access, dangerous script execution, and delayed side effects caused by agent workflows.
Prompt filters inspect the prompt. Gensee Crate uses defense in depth across the full action path: user intent, agent intent, MCP and tool calls, memory, skills, files, network, processes, and execution.
Gensee Crate traces lineage across sessions. If a risky action is caused by earlier memory writes, skill changes, generated artifacts, or local context, Crate can connect the final action back to the earlier cause.
Gensee Crate treats memory writes and reads as policy surfaces. It can trace when a memory was planted, when an agent later used it, and whether resulting actions should be allowed, denied, escalated, or reported.
Yes. Gensee Crate is designed as a low-latency sidecar beside unmodified agents. It can run beside tools such as Claude Code on endpoints like MacBook without requiring teams to rebuild their agent stack around a new SDK.
Developer (Open Source) is planned for individual developers on a single trusted endpoint. Enterprise is for company-managed environments with company-set policy, distributed enforcement, on-prem evidence, central audit, and integrations with internal systems.
Yes. The enterprise product is designed for on-prem or customer-cloud deployment so company policy, audit evidence, traces, and incident data can stay inside the customer's environment.
Enterprise deployments can integrate with identity systems, SIEM, Slack or PagerDuty, code hosts, MCP gateways, internal developer platforms, policy systems, and evidence or incident workflows.
Preliminary AgentCanary Benchmark results show Gensee Crate improving defense rates across memory poisoning, long-horizon tasks, and prompt injection. The main Gensee Crate page reports the latest tested rates and notes the local test setup used for those results.
Talk to us about endpoint sidecars, company policy, on-prem evidence, and how Gensee Crate would fit your agent environment.
Book a demo