NEWv0.1 · LangGraph + Claude · Read the architecture

Build it once. Agents run it.
Governance keeps a human on the risky steps.

Minctrl is the AI-native workflow builder for regulated operations. Design a process once; AI agents execute it end-to-end, and a governance layer — risk-tiered gates, a panel of compliance judges, calibrated confidence and a tamper-evident audit trail — decides exactly when a step pauses for human sign-off.

See every decision an agent made, why each gate paused, and who signed off — in a tamper-evident trail your auditor can read.

LangGraphClaude OpusClaude SonnetTemporalPostgresBPMN 2.0PythonTypeScript
minctrl · runs · aml-case-A4471
LIVE
Run — AML alert triage
0.0s
01
Ingest alertIntake · auto
02
Enrich & screen
03
Risk assessment
04
SAR-filing gate
GATE
05
File SAR
06
Close & seal
streaming consolesteps · 1
[00:00] ▸ alert ingested · case A-4471 · $82,400 wire
Step
Ingest alert
Confidence
…
Human gates
1
Audit
live

Minctrl turns a regulated process into a running, governed workflow — 45 ready across fintech, telecom, health, legal & insurance, or describe your own.

45
Governed flows
11
Verticals
~582
Governed steps
BPMN 2.0
Import + export
Temporal
Durable park/resume
tamper-evident
Audit trail
Built on
psychology
Anthropic Claudeprimary LLM
device_hub
OpenRouterfallback provider
schema
LangGraphagent orchestration
storage
PostgreSQL + pgvectorknowledge store
cloud
Cloudflare Pagesedge hosting
verified
ISO 20022 + CBPR+Nov 2025 ready
Frameworks the governance layer maps to
EU AI Act →DORA →MiCAMiFID IIAML5/6GDPRNIS2PSD2SOC2ISO27001BaFinFCATFReIDAS 2
Why it's different

Agents build it.
Governance decides who signs off.

Governance built in, not bolted on

Every risky step is a gate. A tier resolver routes it — low-risk reversible work can auto-clear, medium goes to a panel of compliance judges, irreversible always waits for a human. Calibrated confidence and a tamper-evident audit trail on every decision.

Deterministic process, not a black box

Design → validate → run → gate → park → resume → audit. Every run follows the same replayable path with durable checkpoints — parked runs survive restarts and resume exactly where a human signed off.

Domain expertise for regulated ops

Operational-process depth, not generic automation: AML/SAR triage, prior-authorization, chargeback representment, collections under FDCPA — 45 flows across 11 verticals, each with the right gates pre-wired.

What you get

Everything a governed run needs,
in one platform.

Design, run, gate, and audit — plus the connectors and BPMN interop to fit your existing stack.

Visual process canvas

Design a process as a node graph, then validate it — reachability and gate-removal checks before anything runs.

45 governed flow templates

Real branched processes across 11 verticals, each with the right risk gates already wired in.

Risk-tiered gates + judges

Every step scored by blast radius and reversibility. A compliance-judge panel weighs in on the medium-risk ones.

Configurable connectors

Bind any step's tool to your own HTTP API — base URL, method, auth, field mapping — and test the wiring before you save.

BPMN 2.0 import / export

Bring an existing BPMN model in, or export any Minctrl process out. Round-trips with your BPM tooling.

Durable park / resume

A Temporal-backed runtime parks a run at a gate and survives restarts — resuming deterministically on sign-off.

Tamper-evident audit trail

Who decided what, why each gate paused, which connector fired — sealed into the artefact a regulator asks for.

Calibrated confidence

Each agent decision carries a calibrated confidence score, so a low-confidence step can escalate to a human.

Under the hood

Deterministic. Auditable.
Design once, run governed.

Design a process once — on the canvas or from your existing BPMN — then watch a governed run drive it step-by-step, calling real APIs through connectors and parking for human sign-off exactly where risk demands. Durable, replayable, sealed.

Stage 1 · Design

Lay out the process on a node canvas — or import your existing BPMN 2.0. Every step is a typed node; bind its tool to a real HTTP connector.

Canvas · Process Cockpit
Inputs
→ Process goal
→ Existing BPMN 2.0 (optional)
Outputs
← Node graph on canvas
← Steps bound to connectors
← Gates placed on risky steps
Runs on
Process Cockpit canvas
Phase
authored
in the run lifecycle
Judgement agents
Claude Opus

Deep reasoning at the gate: risk-tiering, judge panel, calibrated confidence on irreversible steps

Gate resolverCompliance-judge panelRisk assessor
Step agents
Claude Sonnet

Per-step execution across 45 flows — enrich, screen, decide, call the bound HTTP connector

Step executorEnrichmentCase analystConnector caller
Runtime agents
Claude Haiku

High-frequency deterministic work: intake, routing, sealing the audit trail, replay guards

IntakeRoutingAudit sealerIdempotency guard

Need the underlying app too? The same agents can scaffold the vertical application a process runs on — see the AI app builder.

Governance · human-in-the-loop

Every step, risk-tiered.
You choose the autonomy.

Minctrl scores every step by blast radius and reversibility, then routes it: low-risk reversible work clears automatically, medium-risk goes to a panel of compliance judges with calibrated confidence, and irreversible actions always wait for a human. A tamper-evident audit trail records who decided what — the artefact a regulator asks for.

Not every step stops. The resolver routes each one by risk — reversible work clears on its own, and an irreversible step (filing a SAR, releasing a payment) can never fire without a human sign-off.

Start
Ingest & enrich
Risk assessment
judge panelPAUSE
Sanctions screen
judge panelPAUSE
SAR filing
human sign-offMANDATORY
Payment release
human sign-offMANDATORY
Close & seal audit
Done
process step (auto-clears when low-risk)judge panel (medium-risk)human sign-off (irreversible)
WHAT WE AUTOMATE

45 governed flows, ready to run

These aren't demos. Each is a real, branched process — multi-step, gated at every irreversible action — that runs across fintech, telecom, health, legal, insurance and more.

See a governed flow
Intake → Screen → Risk gate → Execute → Sealed audit
illustrative
Intake
1 agent · low risk · auto
Screen / Enrich
2 agents · parallel
parallel
Risk gate
GATE · human sign-off
Execute
1 agent · high risk
irreversible
Sealed audit
done · audit sealed

Reversible steps run on their own; the risk gate parks the run and waits for a human before the one irreversible action — then everything is sealed into a tamper-evident audit trail.

low riskmedium riskgate · humansealed / done
KYC/AML compliance autopilot
15 steps · 1 gatebranched
Chargeback representment autopilot
21 steps · 4 gatesbranched
Complaints handling & redress autopilot
13 steps · 1 gatebranched
Document & liveness verification autopilot
13 steps · 1 gatebranched
DSAR & privacy-request fulfilment autopilot
13 steps · 1 gatebranched
Failed-payment smart-dunning autopilot
16 steps · 2 gatesbranched
Cash-flow underwriting autopilot
14 steps · 2 gatesbranched
Payment exception repair & STP auto-repair autopilot
13 steps · 2 gatesbranched
Transaction-reporting exception remediation autopilot
13 steps · 2 gatesbranched
Stipulation & condition-clearing autopilot
12 steps · 1 gatebranched
Tier-1 card & account servicing autopilot
12 steps · 1 gatebranched
Network alarm correlation autopilot
16 steps · 2 gatesbranched
B2B churn save-desk autopilot
12 steps · 1 gatebranched
Bill-shock dispute resolution autopilot
12 steps · 2 gatesbranched
Inter-carrier settlement reconciliation autopilot
13 steps · 1 gatebranched
Network incident triage & routing autopilot
13 steps · 1 gatebranched
Order fallout auto-diagnosis & remediation autopilot
14 steps · 1 gatebranched
Revenue-leakage recovery autopilot
13 steps · 1 gatebranched
SIM-swap / port-out fraud-defense autopilot
9 steps · 1 gatebranched
Truck-roll avoidance autopilot
16 steps · 3 gatesbranched
Tier-1 care deflection autopilot
13 steps · 1 gatebranched
Bookkeeping & close autopilot
13 steps · 1 gatebranched
SOX ITGC audit autopilot
13 steps · 1 gatebranched
Debt-collection / AR autopilot
15 steps · 1 gatebranched
Mortgage-underwriting autopilot
13 steps · 1 gatebranched
Tax-prep autopilot
12 steps · 2 gatesbranched
Provider-credentialing autopilot
11 steps · 2 gatesbranched
Clinical-trial-ops autopilot
11 steps · 1 gatebranched
Pharmacovigilance autopilot
13 steps · 1 gatebranched
Prior-authorization autopilot
15 steps · 2 gatesbranched
Medical-coding autopilot
12 steps · 1 gatebranched
Estate-planning + probate autopilot
14 steps · 1 gatebranched
Immigration-petition autopilot
12 steps · 1 gatebranched
Legal-document autopilot
12 steps · 2 gatesbranched
Patent-prosecution autopilot
11 steps · 1 gatebranched
Title & escrow autopilot
11 steps · 1 gatebranched
Real-estate appraisal autopilot
11 steps · 1 gatebranched
Claims & underwriting autopilot
12 steps · 1 gatebranched
Workers'-comp claims autopilot
13 steps · 1 gatebranched
Customs-clearance autopilot
11 steps · 1 gatebranched
Freight-brokerage autopilot
13 steps · 1 gatebranched
Managed-IT autopilot
13 steps · 2 gatesbranched
Payroll-processing autopilot
11 steps · 1 gatebranched
Source-to-pay autopilot
13 steps · 1 gatebranched
Managed-SOC / MDR autopilot
11 steps · 1 gatebranched
Full transparency

Every step. Every gate.
Every decision logged.

Integrated views into a governed run — the live console, a tamper-evident audit trail, the execution graph, plus the board, wiki and analytics.

Watch a governed run, live
Dashboard
Dashboard

Project list + create workflow.

Kanban board
Kanban board

Auto-populated tickets per gate.

Wiki
Wiki

Auto-generated docs across runs.

The money case

Seconds of tokens.
Not analyst-hours.

A governed AI run completes a process step in seconds for cents of tokens — a human analyst takes minutes to hours at a loaded rate. The human stays only on the irreversible step.

~$0.03–0.30
Cost per governed run

cents of LLM tokens per case, vs analyst-hours

seconds
Time per case

an agent runs the reversible steps end-to-end

~1 in N
Human-touch reduction

a person reviews only the irreversible gate, not every case

Worked example · AML alert triageillustrative — not customer data
Manual analyst
Minctrl governed run
Time per case
Manual analyst~20 min triage
Minctrl governed runseconds of agent time
Cost per case
Manual analyst~$25 (at ~$75/hr loaded rate)
Minctrl governed run~cents of LLM tokens
Human touch
Manual analystevery case, end to end
Minctrl governed runonly the irreversible gate (e.g. SAR filing)
Audit trail
Manual analystnotes + email approvals
Minctrl governed runsealed, tamper-evident, per run

Figures above are illustrative examples for shaping intuition — not measured customer results. Token cost varies with model tier, case complexity and prompt size; analyst time and loaded rate vary by team and jurisdiction.

For developers

Build on the API.
Auth → connect → run → resume.

Every governed run is an API call away. Authenticate, bind a step to your own HTTP API, start a run that parks at the first human gate, then record the decision to resume it.

1 · Authenticateregister or log in → bearer token
curl -X POST https://api.minctrl.com/auth/login \
  -H "Content-Type: application/json" \
  -d '{"email":"you@bank.com","password":"••••••"}'
# → { "token": "eyJhbGciOi..." }   (also: POST /auth/register)
2 · Bind a connectorpoint a step's tool at your own HTTP API, then test it
curl -X PUT https://api.minctrl.com/connectors/sanctions_screen \
  -H "Authorization: Bearer $TOKEN" -H "Content-Type: application/json" \
  -d '{"base_url":"https://api.yourbank.com","method":"POST",
       "path":"/screen","auth_type":"bearer","auth_token":"$KEY",
       "body_map":{"name":"customer.name"},"reversible":true}'

curl -X POST https://api.minctrl.com/connectors/sanctions_screen/test \
  -H "Authorization: Bearer $TOKEN" -H "Content-Type: application/json" \
  -d '{"config":{...},"state":{"customer":{"name":"Acme Ltd"}}}'
3 · Start + resume a runstarts governed, parks at the first human gate, resume records the decision
curl -X POST https://api.minctrl.com/process-runs \
  -H "Authorization: Bearer $TOKEN" -H "Content-Type: application/json" \
  -d '{"vertical":"kyc-aml","inputs":{"case":"A-4471"}}'
# → { "run_id": "...", "status": "parked", "gate": "signoff" }

curl -X POST https://api.minctrl.com/process-runs/$RUN_ID/resume \
  -H "Authorization: Bearer $TOKEN" -H "Content-Type: application/json" \
  -d '{"gate":"signoff","decision":"approved"}'

curl https://api.minctrl.com/process-runs/$RUN_ID \
  -H "Authorization: Bearer $TOKEN"     # full run + sealed audit trail
4 · Templates & BPMNlist templates, fetch the canvas graph, export BPMN 2.0
curl https://api.minctrl.com/process-templates/ \
  -H "Authorization: Bearer $TOKEN"

curl https://api.minctrl.com/process-templates/kyc-aml/canvas -H "Authorization: Bearer $TOKEN"
curl https://api.minctrl.com/process-templates/kyc-aml/bpmn   -H "Authorization: Bearer $TOKEN"

The full OpenAPI / Swagger schema is served by the API at /docs — every endpoint, model and response documented and try-it-live.

Pricing

Pay per run.
Not per seat.

Three tiers — start free, pay only when a governed run executes. Enterprise gets bespoke compliance frameworks and on-prem deploy.

Free
$0forever

Explore governed automation. Bring your own LLM key.

  • 1 governed run per day
  • All 45 governed process flows
  • Tamper-evident audit trail per run
  • Best-effort support
Recommended
Pro
$49per month

For teams running regulated operations. Priority queue.

  • 50 governed runs / month
  • Private repos + branch protection
  • All 14 compliance frameworks
  • Audit-pack generation included
  • Slack notifications
  • Email support
Enterprise
Customtalk to founder

For regulated operators and licensed fintechs. SLA + dedicated infra.

  • Unlimited governed runs
  • On-prem / VPC deployment
  • Custom domain agents
  • DORA / SOC2 / ISO27001 attestation
  • 99.9% uptime SLA
  • Dedicated success manager

Avg cost per governed run: cents in LLM tokens · Free tier covers your first 30 runs

FAQ

Real questions from real CTOs.
Asked before they signed up.

You do. The full output — source, tests, IaC, audit pack, wiki — is yours under MIT-style ownership with no proprietary watermarks. You can fork, modify, and redistribute. The code is committed to a git branch on your GitHub repo (or downloaded as a zip if no repo is connected). Minctrl claims no IP rights to anything we generate for you.
Have a question we haven't answered? hello@minctrl.com — typical response within 24h.
45 governed flows, ready to run

Design it once.
Agents run it. You sign the risky steps.

Start a governed run in minutes — agents execute end-to-end through your own APIs, park at the gates you define, and seal a tamper-evident audit trail. For cents per run.

✓ No credit card✓ Tamper-evident audit trail✓ Human sign-off on irreversible steps