Identify
We'll do the audit with you in the demo — surface the recurring work that's draining your senior team.
Solution: Back-office automation
Dunning, vendor invoice triage, expense classification, recurring email drafting — handled by agents that wait for your approval whenever it matters. The work that takes a Senior AM half their week; runs as an agent stack with HITL where you set the gates.
When the recurring work that drains your back-office team is mechanical, you want agents to handle it with HITL where it matters, so that your headcount stays focused on what only humans can do.
How it works
We'll do the audit with you in the demo — surface the recurring work that's draining your senior team.
Pick from the 50+ template library or compose new agents. Attach an SoD policy that mirrors your existing approval gates.
With your approval gates set where you want them. Every action enters the hash-chained ledger; signed evidence pack on demand.
Capabilities
Detects failed payments via Stripe webhooks, sequences follow-up notifications, escalates after N days. Indefinite-retention audit row per dunning event.
Classifies Mercury statement transactions, flags anomalies, queues material items for SoD-policy approval. Detection thresholds are named constants in the vendor-watchdog service.
CFO sub-agent classifies expenses against budget categories; flags exceptions for Controller review. Material actions never bypass an approver.
Drafts under HITL — never sends without approval. Same primitive that backs the support-agent reply flow today.
Use cases
Stripe webhook fires; agent sequences notifications; HITL gate on escalation.
Anomaly flag on amount jumps, unknown vendors, off-pattern timing.
Classification + budget-category match + Controller approval queue.
Filters inbound asks against your priorities; drafts holds; you confirm.
Surfaces upcoming renewals from Mercury transaction history; flags ones that need a decision.
Outgoing COO responsibilities replaced with agents at LMNTL
Playbook → agents transition timeline
Production agents running LMNTL itself
| Capability | GoodHelp | DIY Python + LangGraph | Hiring a junior |
|---|---|---|---|
| Setup time | Hours-days | Months | Weeks (recruit + onboard) |
| Ongoing cost | Per-agent fee + LLM passthrough | Engineering team | $50-80k+ fully loaded |
| Audit trail | Built-in | DIY | Manual log |
| Approval gates | Built-in | DIY | Human-mediated |
Comparison reflects publicly documented capabilities as of 2026-05-25. Customer mileage may vary by configuration.
No. HITL is the default. The customer-facing email cadence is delegated to Stripe-hosted dunning + Smart Retries; the agent's job is to detect, sequence, and surface — never to send without your approval on escalation steps. Configurable per agent.
Yes. Hash-chained action ledger captures every action; signed evidence pack — JSON manifest plus body, KMS-rooted signing key — on demand.
Approval gates catch high-risk actions before they happen. Lower-risk actions can be replayed or corrected via the ledger — the action is recorded and reversible at the ledger layer.
Per-agent + LLM passthrough. See the pricing page.
Thirty minutes — bring one recurring task that drains your team; we’ll show you what an agent stack looks like with the approval gates set where you want them.