Add Cordum — runtime safety kernel for AI agent orchestration #11
Description
Project
- Name: Cordum
- Repository: github.com/cordum-io/cordum
Description
Cordum is a safety-first agent orchestration platform with AI safety as a core design principle:
- Safety-before-dispatch — gRPC safety kernel evaluates every agent action against configurable policies before execution. No tool runs without policy clearance
- Output scanning — 2-phase output policy: scan agent output for PII, secrets, prompt injection before it reaches users. Quarantine or redact unsafe content
- Escalation flows — policies can require human approval before execution (approval gates)
- Audit trail — every policy decision, tool call, and state transition is recorded with full context
- Fail-closed design — invalid config or policy errors prevent service startup, never fall back to permissive defaults
Open-source, self-hostable. 7 Go services.