Catch AI agent failures that look like success.

Detect duplicate actions, conflicting updates, and false success claims across your business systems. Start in watch-only mode in minutes.

Book a 15-min Hardening Call

Starts watch-only (never blocks day 1) · Works with LangGraph, Temporal, OpenAI Agents & MCP

</> agent.refund()stripe · shopify · erprepeat · stopped
Built for the systems you can’t afford to break
  • Stripe Refunds
  • Shopify Orders
  • NetSuite / QuickBooks
  • Zendesk / Gorgias
  • Salesforce / HubSpot
How it works

From agent call to proven result.

Agents5 running1 needs youLumni onLast 30m
−25m−20m−15m−10m−5mnow
Support agent
LangGraph
Billing agent
OpenAI Agents
Pricing agent
Temporal
AnaEditing
Human rep · Support
Retention agent
MCP
Ops agentNeeds you
Claude Code
ZReply to Zendesk #1177
SStripe refund $840 · Order #4821
QBInvoice run
SStripe refund $840 · Order #4821
PGWeekly margin report
PGSet Pro plan to $45
AEditing Gorgias ticket #1183
HSChurn-risk scan
GCredit note · Gorgias #1183
SFSync CRMDidn’t happen
NSPay NetSuite AP invoice #9021 Needs you
1. Stops a repeat

Two agents, one refund. Only one reaches Stripe.

01
See it work

Sample agents make real mistakes in a sample workspace, and you watch Lumni catch them.

02
Connect your agent

In its folder. Lumni only watches: you see what it would stop, and nothing is blocked.

03
Protect

When the record shows it only stops real problems, turn it on per action.

Settings → Protect
What you get

The dangerous parts of agent actions, handled.

Retries, stale reads, people and agents on the same record, tools that say OK when nothing happened. Plain rules catch each one before it costs you.

S
Stripe refund $840 · Order #4821
Support agent · LangGraph · 2:14:09
checking…
S
Stripe refund $840 · Order #4821
Billing agent · retry · OpenAI Agents · 2:14:11
checking…
$840 kept · Stripe never saw the copy

Stops the repeats

Retries and second agents reach for the same refund. The first goes through; the copy never reaches Stripe.

price_list · pro_planv41
Read price listv41
Changed by Danav42
Re-read, then write $45v42

Makes it re-read

Data changed after the agent looked. It reads again before it writes.

A
Gorgias ticket #1183
Human rep editing
Retention agentwaiting its turn

People go first

A rep is in the ticket. The agent waits, then goes on the fresh version.

agent-approvals
Lumni 2:31 PM
Stripe refund $1,200 · Order #5530 is over your $500 rule.
ApproveReject

Approvals in Slack

Risky actions wait for a yes, with what you need to decide.

Agent says: “Invoice #9021 posted ✓”
NetSuite reads back: no such invoice
Half-finished ERP write · flagged as an incident

Read-back, not trust

“Done” is checked in the app itself. Half-finished work becomes an incident.

Live demo

The real console. Click anything.

A sample company’s day with Lumni, every screen live. Nothing here touches a real account.

Open full screen
app.lumniverse.com/todaydemo
Loading the live demo…
One call, before the action

It’s just your code.

One check in front of the refund, the email or the CRM write. Plain, deterministic rules answer it. No model decides, so it adds no LLM latency.

refund.tsSDK docs
1const d = await lumni.checkAction({
2 actor_id: 'support-agent', work_id: ticket.id,
3 operation: 'stripe.refund',
4 effect_key: `refund:${order.id}`,
5});
6if (shouldExecute(d)) {
7 await stripe.refunds.create({ charge });
8}
second call · same effect_keystopped
{
  "decision": "stop",
  "reason": "same effect already ran",
  "effect_key": "refund:4821",
  "first_seen": "2:14:09 PM",
  "decided_by": "rule, no model"
}
Watch-only workspaces record this and let it run.
No AI decides

The same effect twice, data that changed, someone on the record, or a rule you set.

TypeScript, Python, Go

Adapters for LangGraph, OpenAI Agents, Temporal and MCP.

Open SDKs

The SDKs, the CLI and the skill are Apache-2.0.

Each line is a decision from a plain rule, made before the action reached a real system.
Sample decisions · hover to pause
SStripe refund $840 · Order #4821
Retry after timeoutRepeat stopped · $840 kept
NSNetSuite AP invoice #9021
Agent said doneHalf-finished · flagged
GGorgias ticket #1183
Human rep editingAgent waited its turn
ShShopify order #10342 · cancel
Order changed after readRe-read before write
HSHubSpot deal #7731 · stage
Rep edited 40s earlierOverwrite blocked
SStripe refund $840 · Order #4821
Retry after timeoutRepeat stopped · $840 kept
NSNetSuite AP invoice #9021
Agent said doneHalf-finished · flagged
GGorgias ticket #1183
Human rep editingAgent waited its turn
ShShopify order #10342 · cancel
Order changed after readRe-read before write
HSHubSpot deal #7731 · stage
Rep edited 40s earlierOverwrite blocked
QBQuickBooks bill #3317
Second payment runPaid once
SStripe refund $1,200 · Order #5530
Over the $500 ruleApproved in Slack
SFSalesforce opp · Q4 renewal
Two agents, one recordOne at a time
ZZendesk ticket #7781
Second agent replyingJoined the first reply
ShShopify refund $64 · Order #4790
Read back from ShopifyProven
QBQuickBooks bill #3317
Second payment runPaid once
SStripe refund $1,200 · Order #5530
Over the $500 ruleApproved in Slack
SFSalesforce opp · Q4 renewal
Two agents, one recordOne at a time
ZZendesk ticket #7781
Second agent replyingJoined the first reply
ShShopify refund $64 · Order #4790
Read back from ShopifyProven
What Lumni catches

Twelve ways agents go wrong. One place that handles them.

Doing it twice
Same payment, twiceLets one through

A retry after a timeout refunds again.

Same email, two answersJoins the first

Two agents reply to one customer.

Paying to redo workStops it before it spends

Model calls for work already done.

Acting on old information
Writing a stale valueMakes it re-read

Acts on data someone just changed.

Plans that changedStops stale work

The customer changed their mind; the agent didn’t.

Out of orderWaits its turn

Ticket closed before the refund landed.

Getting in each other’s way
Agent vs. personYou always win

You fix a record; an agent overwrites it.

Five on one recordOne at a time

Many agents edit one account at once.

Opposite goalsWaits, then re-checks

One agent cancels while another upgrades.

Saying “done” when it isn’t
Never happenedProves it in the app

The tool said OK. The app has nothing.

Half doneFlags it as an incident

Charged the card; the deal never closed.

Runaway swarmSlows it at your limit

One task fans out to forty agents.

After 48 hours in watch-only

See what it would have caught. Then decide.

Watch-only never blocks. It records every action it would have stopped, and after two days you get a risk report with dollar amounts, systems and the exact records.

  • Every repeat, stale write and collision, with what it would have cost
  • Which actions to protect first
  • Something you can forward to your CFO
Watch-only risk report
sample
3
duplicate payment retries caught
$1,840 saved
6
human-rep overwrites it would have blocked
Gorgias and HubSpot
2
half-finished ERP writes flagged
NetSuite, read back
Top findings
  • Stripe refund $840 · Order #4821
    Retry after a timeout
    Would stop: repeat
  • Gorgias ticket #1183
    Human rep editing
    Would hold: person first
  • NetSuite AP invoice #9021
    Agent said done; not in NetSuite
    Flagged: half-finished
  • Shopify order #10342 · cancel
    Order changed after read
    Would re-read: stale
Next: protect stripe.refund and netsuite.invoicenothing was blocked
Pricing

Priced for agents that move money.

Start free in watch-only. Pay when Lumni protects live actions, or have us wire it into your stack and prove it on your own traffic in week 1.

Watch-only

See what your agents would break.

$0/mo
1,000 events/mo included
  • 1,000 events/mo
  • Watch-only: records, never blocks
  • SDKs (TypeScript, Python, Go) + MCP gateway
  • Coding-agent skill
  • The full console and action record

Production

Protect live money-moving agents.

$299/mo
+ $0.003 / event up to 1M, then $0.002 / event
Everything in Watch-only, plus
  • Pay per event, cheaper past 1M
  • Stops repeats, stale writes and collisions
  • Slack approvals for risky actions
  • Read-back verification in the app
  • Unlimited people and agents

Enterprise

We wire Lumni into your stack in 48 hours.

Custom
Pricing tailored to your team
Everything in Production, plus
  • Hands-on integration: LangGraph, Temporal or MCP
  • 48-hour Shadow-Mode Risk Report
  • Custom read-back verifiers: Stripe, Shopify, ERP, CRM
  • A dedicated Slack channel with us
$0 if we don’t catch a real bug in week 1.

One event is one action an agent asks about before doing it: a refund, an email, a CRM write. Reads, dashboards and people using the console are free.

Running it on your own servers? The source is on GitHub · self-hosting guide

FAQ

Things CTOs ask before they turn it on

Didn’t find yours? Book 15 minutes or write to us.

No. New workspaces only watch: Lumni records what it would have stopped and lets everything through until you turn protection on, per action.

Let your agents act.
Once.

Wiring it into a LangGraph or Temporal stack? Book 15 minutes and we’ll look at your riskiest action together.

Anurag
Founder, Lumni