VeriSwarm is a hosted control plane for AI agent trust and governance. Your agents run on your infrastructure. VeriSwarm handles scoring, policy decisions, security scanning, identity verification, and audit logging.
Trust scoring, policy tiers, and moderation decisions. Gate is the foundation — always on, available on all plans. Scores are computed from behavioral events your platform submits. Docs
Security scanning and PII protection for AI agents. Guard detects sensitive data in event streams and tool calls, replaces it with reversible tokens, and provides a kill switch for immediate agent revocation. Guard Proxy sits between agents and their tools, applying PII filtering, policy enforcement, and injection scanning transparently. Docs
Verified agent identity. Ownership claims, signed capability manifests, and scoped delegations establish that an agent is who it claims to be and is authorized to act within defined boundaries. Docs
Immutable audit ledger. Every meaningful action is recorded in a tamper-evident, append-only ledger with cryptographic integrity verification. Export for compliance submissions. Docs
Conversation engine and LLM routing. 14-step pipeline with knowledge retrieval, Guard PII tokenization, complexity-based routing across multiple providers (OpenAI, Anthropic, Google, Azure, Mistral), automatic fallback chains, semantic cache, prompt/context compression, grounding checks, and Vault audit logging. Docs
Agent deployment and lifecycle management. Pre-built templates, knowledge bases, widget chat embedding, integration management, and conversation logging — packaged so every deployed agent has the full trust stack active by default. Docs
YAML-driven intelligent automation. 14 step types (LLM, adapter, condition, transform, wait, human review, webhook, Guard scan, knowledge, platform query, score, send email, Vault log, web search). Triggers via cron, webhook (HMAC), VeriSwarm event, or manual. Pre-step + post-step budget enforcement; every step is Vault-logged. Free=1 active workflow, Pro=10, Max/Enterprise=unlimited. Docs
VeriSwarm exposes its full API via the Model Context Protocol, compatible with Claude Desktop, Cursor, and any MCP client. Docs
Agents can request signed credentials containing their current trust scores. Any system can verify these using VeriSwarm's public key endpoint (/.well-known/jwks.json), enabling trust to travel with the agent across platforms. Docs
Custom governance rules using the Cedar Policy Language. Enterprise and Max plan customers can define declarative policies that govern agent behavior — what actions each trust tier can perform, which resources they can access, and under what conditions.
Cedar policies replace the built-in policy matrix. Free and Pro plans use the default policy (equivalent to the built-in matrix). All policy changes are Vault-logged for audit compliance. API Docs
VeriSwarm maps to 8 frameworks. Three are counsel-reviewed; five are technical_preview:
Counsel-reviewed:
Technical preview (mapped from live posture; not yet counsel-reviewed):
Per-tenant compliance reports available via GET /v1/compliance/{framework}. List frameworks via GET /v1/compliance/frameworks. All policy changes are Vault-logged. Compliance page
Three-state circuit breakers (closed/open/half-open) per LLM provider with configurable thresholds. Tenant-scoped SLO targets (availability %, P95 latency), error budget tracking with burn rate projection, and an SRE dashboard combining all signals. API Docs
Topic trending, knowledge base coverage metrics, and knowledge gap detection — observability into what agents know and where they struggle. Identifies agents with high failure rates and low KB coverage, ranked by gap severity with recommendations. API Docs