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 agents are organised around Skills. A standard skill library ships out of the box, organisation-specific tasks are added through the Agent Builder, and the system also proposes new skills automatically.

​
What a skill is

A skill is a single unit of work an agent can perform. Examples:
  • Generate a weekly sales report
  • Build a quotation PDF
  • Draft an inquiry email to a vendor
  • Render a chart from a data set
  • Call an external API
Each skill defines its trigger (when to fire), inputs (what info it needs), behaviour (which tools to chain), and output (the response shape).

​
Proactive skill generation

When the system notices employees doing the same task repeatedly, it proposes a new skill. 
Every Monday at 09:00 KST
  ↓
1. Collect Slack conversations + tool execution logs from the last 7 days
  ↓
2. Use Gemini to extract 5 candidate repeating patterns
  ↓
3. Dedupe against existing tools and skills (keyword + LLM semantic compare)
  ↓
4. Slack DM to admins with [Approve] [Reject] buttons
  ↓
5. On approval: skill activates immediately (no redeploy)
Dedup safeguards: 90-day cooldown for rejected patterns. Skills semantically similar to existing ones are dropped automatically.

​
Agent Builder

Reachable at /admin/agents/*, the Agent Builder UI lets you:

Define agents

Edit role, tone, behavioural rules, and the system prompt.

Add skills

Create new skills from a template, or clone and edit existing ones.

Activate tools

Pick which tools from the standard library this agent can use.

Live test

Chat with the agent inside the builder. No redeploys; iterate freely.

​
Standard tool library

Categories of built-in tools every agent can opt into:
CategoryExamples
FilesExcel/CSV/PDF/HWP/PPTX/DOCX parse and generate, charts, diagrams, barcodes
SearchWeb search, Naver search, fetch URL, YouTube transcripts, internal business context
EmailGeneral send, personal Gmail read/send (OAuth)
Google WorkspaceDrive, Calendar, Gmail (personal OAuth + service account)
Microsoft 365OneDrive, Outlook, Teams, Calendar
ERP / groupwareDouzone, Wehago (tasks, approvals, messaging)
E-commerceCafe24, Naver Smartstore
Tax / identityHometax, NICE, Korea Customs
NotionPage and database read/write
Browser automationBrowserbase stealth browsing — bot-detection evasion, accessibility-tree refs
WorkflowDemand/sales forecast, threshold alerts, weekly report templates, compliance check
Memory & searchChannel history, business context BM25 search
OpenAPIAuto-generate tools from any external API’s OpenAPI spec
The full tool catalogue is visible from the Tools page in the admin console.

​
Planning mode

For complex tasks (build a website, generate a report, set up a connector), the agent asks clarifying questions before doing the work: 
User: "Make me a weekly sales report"
  ↓
Agent: "What time range? (last week / this month / custom)"
User: "Last week"
Agent: "What output format do you want — PPTX, PDF, or Excel?"
User: "PPTX"
  ↓
[Agent gathers data, builds charts, writes the report]
This prevents long-running tasks from failing on a wrong assumption.

​
Context efficiency

Long conversations stay cheap and high quality thanks to automatic optimisations:
  • Observation masking — tool outputs older than 3 turns are replaced by compressed summaries (~30–40% token savings)
  • Compaction — automatic summarisation when the context window hits 70%
  • Prompt cache — system prompts and tool definitions are LLM-side cached for ~75% input cost reduction
  • Context compression phase — additional compression in the channel command pipeline

​
Confidence gate

When the agent’s confidence in an answer is low, it asks a clarifying question instead of guessing. 
User: "What happened with that thing?"
  ↓
[confidence < 0.20 → clarify mode]
  ↓
Agent: "Which thing — yesterday's report, last week's quotation, or something else?"
This gate is a key defence against hallucination and is on by default for every active organisation.