Impraint Guard
Governance, observability and compliance for AI agents, from a single Python SDK.
Impraint Guard sits next to your AI agents and records what they do, what it costs, and whether it was safe. You add a few lines of code; Guard does the rest.
Beta
Guard is in beta. Gemini (Vertex AI and AI Studio) is fully supported today, including LangChain and LangGraph. Other providers are on the roadmap, see Compatibility.
What Guard gives you
- Tracing. Every agent run becomes a trace made of steps: LLM calls, tool calls, retrievals.
- Cost and token tracking. Input, output, thinking and cached tokens, with cost calculated for you from model pricing.
- Anomaly detection and alerts. Unusual behaviour is flagged and your team is notified by email or Slack.
- Content safety. PII and secret detection and masking, plus checks for prompt injection.
- Rules and circuit breaker. Block, throttle or require human approval before a risky action runs.
- Approvals. Hold a risky action until an admin approves it in Slack or the dashboard.
- Compliance reports. Evidence reports for OWASP and India's DPDP Act, built from what Guard records.
- Fail-open by design. If Guard is unreachable, your agent keeps running.