Signed, enforced, audited.
AgentSight wraps your AI agents with Ed25519-signed policies enforced in Rust before any Python runs. Tool violations are blocked in-process — not logged after the fact.
0
Agents registered
0
Built-in tools
0
Tool calls enforced
0.0ms
Avg enforce latency
Define allowed tools in a manifest. AgentSight signs it with Ed25519 and stores a tamper proof record.
Rust intercepts every tool call before Python sees it. Violations are blocked in-process — not logged after the fact.
Tamper evident hash chain logs every event. SOC-2 ready out of the box with full export support.
Every policy manifest is signed with a hardware grade Ed25519 key pair.
Tool call interception runs in Rust before any Python bytecode executes.
High privilege tool calls require multi factor approval before execution.
Hash chained audit events — any modification of history is cryptographically detectable.
Immutable audit trail, role based access, and policy versioning satisfy SOC-2 controls.
Per tool constraints (min/max, regex, enum) block bad values before the function runs.
Outbound HTTP calls are matched against host/method/path rules. Violations are blocked and audited.
Stream live audit events to your terminal. Filter by outcome, follow in real time with ANSI colors.
Domain level counters for tool calls, blocks, MFA events, and active agents — ready for SLO alerts.
Optionally store actual call parameters with PII masking — expand any audit event for full incident context.
Every tool cryptographically signed. Every agent registered.