This guide is for platform teams integrating VeriSwarm as a trust and moderation layer. It covers API keys, decision checks, event ingestion, reports, and billing.
A VeriSwarm workspace is an account with API and billing capabilities enabled. With it, you can:
Enable provider mode from your account console.
Create and revoke API keys from the account console. Your platform systems should authenticate with API keys, not user login tokens. Keys can be rotated independently without affecting your account.
Best practices:
Decision checks are the core runtime integration — call them before any action that needs a trust gate.
Endpoint: POST /v1/decisions/check
Use before actions like:
The response includes a decision, reason_code, and policy_tier. Handle decisions as follows:
Events are the behavioral input that drives scoring.
Endpoint: POST /v1/events
Send events for meaningful agent activity:
When async ingestion is enabled, scoring completes shortly after the event is accepted rather than inline.
Send structured trust and abuse signals from your own moderation systems via the reports API. Supports both single and batch submission.
Use reports for:
If your platform has its own agent IDs, link them to VeriSwarm identities. This keeps your internal records and VeriSwarm in sync and prevents identity drift.
When registering an agent, store the returned VeriSwarm agent ID alongside your internal record.
The account console shows your current plan, daily quota usage, remaining quota, subscription state, and payment methods.
Available plans: Free, Pro, Max, and Enterprise.
VeriSwarm provides:
At registration: Register each agent in VeriSwarm and store the returned agent ID.
During runtime: Send meaningful events, forward abuse or trust reports from your own systems, and call decision checks before sensitive actions.
During incident response: Inspect recent flags and scores, review timeline history, and decide whether to quarantine, challenge, or restore the agent.
When VeriSwarm is unavailable, apply a safe default posture:
reviewdenyThis matches VeriSwarm's safety bias -- unknown trust states default toward caution, not permissiveness.
Guard Proxy is a transparent MCP proxy that intercepts every tool call between your agents and their MCP servers. It applies policy checks, PII tokenization, prompt injection scanning, and audit logging -- with no agent-side code changes.
The fastest setup. Replace your agent's MCP server URL with the VeriSwarm proxy URL:
# Before
MCP_SERVER_URL=https://my-mcp-server.example.com
# After
MCP_SERVER_URL=https://proxy.veriswarm.ai/v1/guard/proxy/my-mcp-server
Or generate the config from the SDK:
Python:
config = client.guard_proxy_config(agent_id="agt_123", server="my-mcp-server")
Node:
const config = client.guardProxyConfig({ agentId: "agt_123", server: "my-mcp-server" });
For production deployments where traffic must stay in your network:
docker run -d \
-e VERISWARM_API_KEY=vsk_... \
-e UPSTREAM_MCP_URL=https://my-mcp-server.example.com \
-p 8080:8080 \
veriswarm/guard-proxy:latest
Point your agents at http://localhost:8080 (or your internal DNS).
When Guard Proxy is active, the following are enabled by default:
For full Guard documentation including PII rehydration, kill switch, and tool policies, see Guard.