Add Cordum — runtime safety layer for LLM agents #77
Description
Project
- Name: Cordum
- Repository: github.com/cordum-io/cordum
- Suggested Section: Tools
Description
Cordum is a safety-first agent orchestration platform that provides runtime security for LLM agents:
- Input policy evaluation — gRPC safety kernel evaluates every tool call against configurable policies before execution. Actions: deny, escalate, allow. Catches prompt injection, unauthorized tool use, and policy violations at runtime
- Output scanning — 2-phase output policy scans agent responses for PII, secrets, prompt injection patterns before they reach users. Quarantine or redact unsafe output
- Audit trail — every policy decision, tool call, and state transition is recorded with full context
- Multi-tenant isolation — per-tenant, per-agent credential scoping and policy evaluation
- MCP server — native Model Context Protocol support
Defense-in-depth for LLM agent deployments. 7 Go services, self-hostable.