01 · Onboarding
Docs → spec in 3 minutes
Drop a PDF, form screenshot, or FAQ doc. Five Claude stages extract every field, rule, and routing path with a verbatim source quote. No hallucinated rules.
Solutions
Automated approval flows
for any policy doc
AI expense approval that runs everywhere your team submits — web form, email, Slack, or Lark. Kopi checks every submission, verifies invoices, flags anomalies, and enforces your policies before the approver opens the request.
⚡
Under 60-second approvals
Kopi delivers AI check results directly to the approver in Lark within 60 seconds — compared to hours of manual review.
🧾
AI invoice verification
Claude AI vision reads PDF and image invoices, extracts key fields, and cross-validates against the expense form. Learn more about our automatic invoice verification.
🚨
Expense anomaly detection
Unusual amounts, date mismatches, and duplicate submissions are flagged automatically before they reach your queue. See how expense anomaly detection works.
🧠
Rule learning
Every approver decision is saved. The same expense next month is recognized instantly — zero manual review needed.
📋
Policy enforcement
Kopi learns your expense policies and checks every submission for compliance before it reaches an approver.
🔗
Native Lark & Feishu integration
Works on whatever channel your team uses — web form, email, Slack, or Lark. Submitters file as normal; approvers get a three-color AI check card.
How AI expense approval works
Kopi plugs into your existing Lark approval flows — no new tools, no workflow changes for employees.
01
Employee submits in Lark
Your team continues submitting expenses through Lark's built-in approval flow. No new app, no new login.
02
Kopi runs 5-dimension AI check
Kopi verifies the invoice, checks the amount against historical norms, validates dates, matches against your rule library, and checks policy compliance — all in parallel.
03
Approver gets a check card in Lark
A three-color result card (green/yellow/red) appears in the approver's Lark chat. Green = approved. Yellow = needs attention. Red = blocked. One tap to confirm.
04
Rule library learns
Every confirmed decision is saved. Next time the same expense type appears, Kopi recognizes it instantly — reducing manual review over time.
Frequently asked questions
- How does AI expense approval work in Lark?
- Kopi connects to your Lark workspace via webhook. When an employee submits an expense, Kopi automatically downloads attachments, verifies invoices with AI vision, checks against your policy rules, and sends a three-color check card to the approver — all in under 60 seconds.
- What expenses can Kopi automatically approve?
- Kopi can auto-approve expenses that match confirmed rules in your rule library — for example, a recurring vendor your team uses monthly. New or unusual expenses are flagged for human review with a detailed AI analysis card.
- How long does AI expense verification take?
- Kopi delivers AI check results to the approver in Lark within 60 seconds of submission — compared to hours or days with manual review.
- Is Kopi secure for financial data?
- Yes. Kopi processes data on secure infrastructure, never stores invoice images beyond analysis, and uses encrypted connections with Lark's official API.
- How do I set up expense approval rules?
- Rules are learned automatically. Every time an approver confirms a category in the Lark check card, Kopi saves that decision to your rule library. You can also view and delete rules in the Kopi dashboard.
Free for Singapore SMBs in private beta
No credit card. No time limit. Set up in 10 minutes.
Built on Mirror
One agent reads every policy doc — across every domain.
Expense approval is one of 11 domains Mirror handles — same pipeline, same citation contract, same runtime. Hand her your policy doc and the form, rules, and routing land in three minutes.
02 · Runtime
Decisions in 4ms
runTemplate(spec, request) → Decision is pure code, no LLM at decision time. The spec is data, the runtime is deterministic. Three-color verdict with full routing trace.
03 · Channels
Web · Email · Slack · Lark
A submission can come from anywhere. Reviewers see the same triage card across all surfaces. Web + email is the default; Slack and Lark are accelerators for already-connected tenants.
Citation contract — non-negotiable
Every field, rule, and routing approver Mirror emits carries a verbatim quote from your source document. If she can’t cite it, she drops the item. No fabricated policies — your AI is anchored to your own words.
Read the docs →Architecture, full pipeline detail, page map.