Authorization infrastructure for AI actions

One API call between your agent and sensitive systems—programmable authorization and receipts before operational changes run.

AI systems are moving from generating text to taking actions. AI actions require authorization.

APPROVEDREJECTEDREVIEW_REQUIRED

Every AI action should return a verdict.

Add approval checks to AI workflows in minutes
Human review for risky actions
Replayable approval receipts for clients

Add once. Reuse across every client workflow.

Protect your first workflow Try a live verify

Partner intake — API key in minutes

AI agent

BiDigest verify

APPROVEDREJECTEDREVIEW_REQUIRED

CRM · payments · ERP

Returns exactly one verdict per call.

Inline authorization check

Verify the proposed action before CRM, billing, or payments change—then branch on a single verdict.

[ AI Agent Proposes Action ]
             ↓
[ BiDigest verify API ]
             ↓
     One verdict per verify call
             ↓
┌───────────────┬────────────────┬───────────────────┐
│  APPROVED     │   REJECTED     │  REVIEW_REQUIRED  │
└───────┬───────┴────────┬───────┴─────────┬─────────┘
        │                │                 │
   (executes route)  (hard stop)     (pauses canvas)
        │                │                 │
        └────────────────┴─────────────────┘
                         ↓
           [ Receipt logged · receipt_id ]

Developer path: POST /api/v1/admissibility/verify — copy-paste recipes on this site.

Examples: AI actions → verdicts

Train clients to think in operational authorization—not chatbot moderation.

Monitoring is not authorization.

AI action	Verdict
Approve $5,000 refund	REVIEW_REQUIRED
Delete CRM contact	REJECTED
Read FAQ / low-risk lookup	APPROVED
Trigger invoice payment	REVIEW_REQUIRED
Escalate support ticket with billing change	REVIEW_REQUIRED

Why agencies are adding approval infrastructure

AI workflows are moving from generating text to changing records, approving payments, and triggering operational actions.

Clients increasingly ask: “What controls the action?” Agencies that ship authorization look responsible—not reckless.

Monitoring after the fact is no longer enough for higher-trust client deals
Shipping without approval infrastructure will eventually look negligent to procurement

What changes for your agency

Ship faster with reusable approval infrastructure—better client conversations, not another compliance workshop.

Before	With BiDigest
“Trust us, the AI usually behaves.”	Every action returns a receipt and an approval verdict.
Clients fear autonomous workflows	Clients see approval controls and human review paths
Rebuild approval logic per client	Same verify call reused in every client template
Harder enterprise security conversations	Clearer security and operations story for procurement
Ambiguous blame after incidents	Replayable receipts with timestamps and receipt_id

How it works

Your AI workflow proposes an action
Refund, CRM update, payment, claim, or copilot step—whatever you already automate.
BiDigest returns a verdict
APPROVED, REJECTED, or REVIEW_REQUIRED in one HTTP call—before money, records, or approvals change.
You store the approval receipt
Replayable proof for clients, disputes, and enterprise onboarding—show receipt_id in every run.

Works inside workflows built with

HTTP Request, webhook, or REST only—no native n8n/Make/Zapier marketplace apps yet. Open Integration recipes for copy-paste curl.

n8n
Make
Zapier
OpenAI Agents
LangChain
CrewAI
Flowise
Langflow
Retool
HubSpot
Salesforce

Start lightweight

Pick a mode—no policy workshops on day one. Most agencies begin with Observe or Review; add Enforce when clients require hard stops.

APPROVEDREJECTEDREVIEW_REQUIRED

Mode	What happens
ObserveNow	Log receipts only
ReviewNow	Pause risky actions
EnforceOpt-in	Stop on REJECTED

What happens on REVIEW_REQUIRED?

Your workflow pauses until a human decides—same pattern agencies already use for refunds and CRM writes.

REVIEW_REQUIRED
Verify returns a verdict + receipt_id; downstream steps wait.
Notify your team
Webhook, email, or agency console—wire the channel you already use.
Human approves
Operator reviews context; approves or rejects in your tool of record.
Workflow resumes
Continue the automation with the receipt on file for the client.

Real agency use cases

Refund approval bots
Auto-approve small refunds; REVIEW_REQUIRED above your threshold.
AI SDR workflows
Gate outbound sends and CRM writes before they hit production.
CRM enrichment agents
Block destructive updates; approve high-impact field changes.
Support escalation agents
Pause ticket actions that need a human sign-off.
Insurance claim review flows
Receipt every decision for regulated client accounts.
Invoice approval agents
Verify payment and billing actions before execution.

Protected by BiDigest

Trust signals agencies use in decks, workflows, and client QBRs—partner intake unlocks co-branding.

Protected by BiDigestWorkflow trust mark
Approval receipts enabledPer-run proof
Human review enforcedREVIEW_REQUIRED path

Show badges on client-facing flows and SOW appendices
Differentiate retainers with verifiable authorization—not policy slides
Certification-style partner marks via onboarding (not self-serve marketplace yet)

5-minute integration

Copy, paste, verify—then reuse the same verify call in every client template. Not per-client consulting—reusable infrastructure.

[COMING SOON] Typed npm SDK for agencies. Today: copy fetch or curl below (works with any stack).

const res = await fetch('https://bidigest.com/api/v1/admissibility/verify', {
  method: 'POST',
  headers: {
    Authorization: `Bearer ${process.env.BIDIGEST_API_KEY}`,
    'Content-Type': 'application/json',
  },
  body: JSON.stringify({
    jurisdiction: { regionCode: 'us-fed', industrySector: 'financial' },
    intent: {
      agentId: 'agency-agent:client-acme',
      actionType: 'refund_approval',
      proposedPayload: { action: 'refund_approval', amount: 500, clientId: 'client-acme' },
    },
    sessionContext: {
      sessionId: 'session-acme-refund-001',
      userId: 'support_agent',
      sessionStartTime: new Date().toISOString(),
    },
    groundTruthBinding: {
      domain: 'agency-automation-demo.bidigest.com',
      regulatoryId: 'AGENCY_AUTOMATION_DEMO_V0',
    },
  }),
});
const data = await res.json();
// Map API response → agency verdict (see /partners/ai-automation/recipes)

Full HTTP recipes for n8n, Make, and Zapier

Partner onboarding

Get started in under 5 minutes

Connect, pick a workflow type, copy a recipe, run a live verify—no policy mapping on day one.

1. Protect your first workflow — Partner intake provisions your org and API key. Protect your first workflow
2. Pick a workflow type — Refunds, CRM, claims, payments, or internal ops.
3. Copy an integration — curl, n8n, Make, or TypeScript. Open HTTP recipes
4. Run a live verify — Simulator or your stack—every run returns a verdict + receipt_id. Try a live verify
5. Ship to every client — Add the same verify call to every client template; console shows REVIEW_REQUIRED queue. Agency console

Protected workflow templates (roadmap)

[PROPOSAL] Deployable assets agencies can clone—thresholds, starter rules, and sample receipts when packs ship.

Protected AI Refund Agent
Protected CRM Update Agent
Protected SDR Agent
Protected Invoice Approval Flow
Protected Claims Review Flow

Agency console

Integration recipes

No marketplace download—use HTTP Request (n8n), Webhook (Make), or POST (Zapier) with the curl body below.

Open full HTTP recipes (curl, n8n, Make)

Starter rules (included)

Defaults ship in code—customize via verifyIntent thresholdRules. No framework workshop required.

Workflow	Default	Outcome
Refund over $500	Human review before bind	REVIEW_REQUIRED
refund_approval / wire_transfer / coverage_change	Always review	REVIEW_REQUIRED
CRM contact delete	Deny in examples	REJECTED
Policy / FAQ lookup	Auto-approve when API allows	APPROVED

Developer docs

n8n · Make · Zapier · LangChain · OpenAI Agents SDK · CrewAI · Flowise · custom backends

Architecture (one call)

AI agent / workflow
    → BiDigest verify API
    → APPROVED | REJECTED | REVIEW_REQUIRED
    → CRM · ERP · payments · claims

Copy-paste examples

Verify from your backend—map to APPROVED, REJECTED, or REVIEW_REQUIRED.

{
  "admissibility_status": "APPROVED",
  "reason_code": "200_ADMITTED",
  "decision_receipt": { "receipt_id": "rcpt_84721…" }
}

Verify action API: POST /api/v1/admissibility/verify

[COMING SOON] Typed npm SDK for agencies. Today: copy fetch or curl below (works with any stack).

curl -sS -X POST "https://bidigest.com/api/v1/admissibility/verify" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "jurisdiction": { "regionCode": "us-fed", "industrySector": "financial" },
    "intent": {
      "agentId": "agency-agent:client-acme",
      "actionType": "refund_approval",
      "proposedPayload": { "action": "refund_approval", "amount": 500, "clientId": "client-acme" }
    },
    "sessionContext": {
      "sessionId": "session-acme-refund-001",
      "userId": "support_agent",
      "sessionStartTime": "2026-05-17T12:00:00.000Z"
    },
    "groundTruthBinding": {
      "domain": "agency-automation-demo.bidigest.com",
      "regulatoryId": "AGENCY_AUTOMATION_DEMO_V0"
    }
  }'

Demo policy defaults (integrators): agency-automation-demo.bidigest.com · AGENCY_AUTOMATION_DEMO_V0

When the verdict is REVIEW_REQUIRED

Branch on REVIEW_REQUIRED—not only HTTP 403.
Notify a human (Slack, email) with receipt_id.
Continue or stop after human decision.

What BiDigest is not

— Not an LLM vendor or chatbot platform
— Not observability-only or a blended AI safety score
— Not native n8n, Make, or Zapier marketplace connectors (HTTP only today)
— Not day-one governance transformation or policy workshops
— Not kernel or network severance on this page—API verdict branching in your workflow

Enterprise deployments

Same verify API for higher-trust programs—separate enterprise hub for audits and trustee paths (not required for agency HTTP integrations).

Enterprise governance hub · Legacy visibility agency program

Inline authorization check

Examples: AI actions → verdicts

Why agencies are adding approval infrastructure

What changes for your agency

How it works

Your AI workflow proposes an action

BiDigest returns a verdict

You store the approval receipt

Works inside workflows built with

Start lightweight

What happens on REVIEW_REQUIRED?

Real agency use cases

Refund approval bots

AI SDR workflows

CRM enrichment agents

Support escalation agents

Insurance claim review flows

Invoice approval agents

Protected by BiDigest

5-minute integration

Get started in under 5 minutes

Protected workflow templates (roadmap)

Starter rules (included)

Developer docs

Architecture (one call)

Copy-paste examples

When the verdict is REVIEW_REQUIRED

What BiDigest is not

Enterprise deployments