FluxyChat Docs

Use the AI Governance dashboard or /admin/ai-governance/* APIs to maintain a registry before deploying agents to production.

Registry

Asset	Risk tiers	Auto-approve
Models	low / medium / high / critical	Manual approval tracked
Prompts	same	low + medium → `approved`; high/critical → `pending`
Tools	same	`requiresApproval` for high/critical

Pre-deploy evaluation

curl -X POST "$WORKER/admin/ai-governance/evaluate" \
  -H "Authorization: Bearer $ADMIN_JWT" \
  -H "Content-Type: application/json" \
  -d '{"targetId":"gpt-4o-mini","targetType":"model"}'

Returns passed, score, and evidence for auditor packs.

Evidence export

curl "$WORKER/admin/ai-governance/evidence" \
  -H "Authorization: Bearer $ADMIN_JWT" \
  -o ai-governance-evidence.json

Combine with /soc2 SOC 2 evidence export.

Runtime enforcement

Execution-time checks use D1-backed action policies:

GET /enterprise/ai-policies
POST /enterprise/ai-policies/check
GET /enterprise/ai-policies/violations

SDK (dev)

import { createAiGovernance } from "@fluxy-chat/sdk";

const gov = createAiGovernance({ autoApproveTiers: ["low", "medium"] });
gov.registerModel({
  modelId: "gpt-4o-mini",
  provider: "openai",
  version: "2024-07",
  riskTier: "medium",
  allowedUseCases: ["support"],
  approvedBy: "admin@example.com",
});

Production registries should use Worker KV via /admin/ai-governance/*.

AI governance