Guardclaw

The runtime. After the model produces a plan, GuardClaw checks each concrete action the agent takes, including shell commands, file reads, network calls, and tool invocations, against your policy before any side effect happens. Turn on deny-by-default and anything outside your allowlist is refused.

At runtime, while the agent is acting. Code scanning happens earlier, on code at rest. Model filtering happens at the boundary of the model, on text going in and coming back. GuardClaw watches what the agent does once it starts calling tools, which is a separate moment and sits alongside the other two rather than replacing either.

Your code and your prompts stay on your machine, and security decisions are made locally with no network round-trip. Anonymous telemetry is on by default and carries decision counts, threat scores, timing, platform and version. It does not carry prompts, commands, file paths or personal data, and you can turn it off with guardclaw telemetry disable. Pattern updates check on a schedule you set. If you choose to sync receipts, a receipt carries the decision, the score, the tool name and action, and the file path or URL the tool touched. Your file contents and prompt text travel as hashes rather than as content. Free and paid tiers are treated differently for training, and the GuardClaw terms and the privacy notice are the authoritative statement of what is collected and what it is used for.

No. GuardClaw uses compiled pattern matching (Bloom filters, Aho-Corasick, RE2) and rule-based policy. Given the same configuration and pattern set, the same input gets the same decision, and every decision is written to a receipt you can audit.

Every approved action gets a short-lived, signed, single-use token scoped to just that action; replays fail. Every decision is written into a cryptographically linked receipt chain, a tamper-evident record built for compliance and fast incident review.

Checks run in the fast path, locally, typically in well under a millisecond. Security that adds heavy latency gets disabled under pressure, so low-latency enforcement was a design requirement from day one.

No tool does. GuardClaw publishes its rates (97%+ on our built-in corpus) and layers seven controls so a miss at one has another chance to be caught. You keep your security team and threat model. GuardClaw is infrastructure underneath them.