Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.teeem-ai.com/llms.txt

Use this file to discover all available pages before exploring further.

Teeem AI leaves your existing systems in place and connects only the information and execution flow. No new ERP, no data migration. When an employee asks for something in plain language from the messenger they already use, the following six steps unfold inside a single response.
Teeem AI web chat — natural-language request from an employee

​
The six-step flow

1

Entry — request from where the team already works

Slack mentions, KakaoTalk group rooms, web chat, Microsoft Teams, REST calls from internal portals. No new tool to learn.
@Teeem How did revenue look last week? Break it down by account.
Teeem AI identifies the channel, verifies the sender, and forwards the request to the next step.
2

Context — pull scattered company data into the question

Information needed to answer is gathered from across your systems.
  • Knowledge base — internal docs, policies, report templates
  • ERP / groupware — Douzone or Wehago for invoices, accounts, approvals
  • Email / calendar — the employee’s own Google or MS365 (via OAuth)
  • E-commerce — Cafe24, Naver Smartstore orders and inventory
  • Internal apps — App Pack models (orders, inventory, tasks, etc.)
  • Conversation history — earlier context in the same channel or thread
Only what’s needed is pulled. Information that won’t be used in the answer doesn’t leave the source system.
3

Insight — surface the change points

Not just a list of data — Teeem AI surfaces the meaningful change points.Example: “Last week’s revenue was ₩120M, −15% vs the prior week. The biggest impact was account A (−₩30M). Likely cause: shipping delay.”Tools commonly used at this step:
  • Time-series analysis (forecast)
  • Threshold monitoring (alert_monitor)
  • Document comparison (diff_documents)
  • Data analysis (analyze_data)
4

Trust layer — protect sensitive data and stay within permissions

Before a response is composed, the following checks run:
CheckBehaviour
PII masking10 Korean PII types auto-masked before reaching the LLM
Permission checkRBAC + ABAC verify the sender can access the requested data
Prompt-injection defenceHidden injection attempts in external content (email, web, docs) are blocked
Output leak detectionResponses are monitored for PII or secrets escaping
Approval gatesSensitive actions (external sends, payments, bulk downloads) require human approval
The masked PII is restored exactly before the response goes back to the user.
5

Systems — connect to existing systems instead of replacing them

Beyond the answer, the agent acts — but only when permissions allow.
  • Generate a sales report PPTX/PDF and attach it to a Slack channel
  • Draft a quotation in HWP and email it to the account (with approval)
  • Export tax invoices from Douzone to Excel
  • Create a calendar event with a Meet link
  • Add a record to an internal app (e.g. order management)
The key is not replacing systems. Douzone stays Douzone, Wehago stays Wehago, Notion stays Notion. Teeem AI connects them.
6

Records — audit log of who did what for whom

Every step is recorded in a tamper-evident hash-chain audit log.
  • Who requested what
  • Which data was accessed
  • Which permission checks ran
  • Which tools were called and what they returned
  • What response went out (with PII masking statistics)
GET /admin/compliance-report exports the entire record as a single ZIP — suitable for direct submission during regulatory audits or ISMS-P recertification.

​
At a glance

Employee's natural-language request
    │
    ▼
┌─────────────────────────────────────────────┐
│  1. Entry      messenger · web chat · webhook  │
│  2. Context    KB · ERP · mail · apps · history │
│  3. Insight    surface changes · find patterns  │
│  4. Trust      mask PII · check perms · defend  │
│  5. Systems    answer + act (reports, emails)   │
│  6. Records    audit log (hash chain · 5 yr)    │
└─────────────────────────────────────────────┘
    │
    ▼
Channel response + citations + completion notice

​
Three guiding principles

Systems stay in place

No replacements. Existing ERP, groupware, internal systems are connected, not swapped.

Inside permissions only

Data access, tool execution, external sends — all gated by the sender’s permissions.

Recorded without gaps

Every action is logged tamper-evidently. Five years of retention.

​
Next

Agentic workflow inside one response

See the same six steps unfold within a single message, in detail.

Permission model deep-dive

How RBAC + ABAC + IP allowlist + tool gates combine.