Docs

Everything is documented. Most of it is open-source.

Quickstarts, concept guides, SDK references, redteam catalog, Living Cert payloads, and the full agent-attack skill library.

Quickstart

  • Wire the SDK in 5 minutes
  • Self-host with docker-compose
  • Local-first dev loop

Concepts

  • Agent fingerprint
  • Living Cert score model
  • Findings + repro tests
  • Mutual Defense Network

SDK & API

  • @vouch/sdk (TypeScript)
  • vouch-sdk (Python)
  • Public REST API
  • tRPC dashboard surface
  • OTel-compatible ingestion

Redteam

  • Promptfoo plugin catalog
  • Strategy mixers
  • Vouch-AI offensive agent
  • Cost caps + budgets

Firewall

  • Scanner inventory
  • Custom regex bans
  • ML classifier retraining
  • Tool-policy decision points

vouchctl CLI

  • vouchctl verify
  • vouchctl jwks
  • vouchctl cert
  • Authenticated subcommands (next)

Operations

  • RBAC scopes
  • Audit logs (hash-chained)
  • Data retention
  • BYOK + VPC isolation

Skill library

  • indirect_prompt_injection
  • rag_poisoning
  • memory_poisoning
  • mcp_exploitation
  • tool_call_hijack
  • cross_tenant_escape
  • approval_bypass
  • confused_deputy

Browse the source.

The whole monorepo is Apache-2.0 on GitHub.