This page is not marketing. We list what Teeem AI doesn’t do well, scenarios we don’t recommend, and known limits. Setting honest expectations now produces a better long-term relationship.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.
Scenarios we don’t recommend
Replacing real-time transactional systems
Order processing, payment authorisation, trading — anything needing millisecond response times. Teeem AI replies in 2–10 seconds.
Legal advice
Teeem AI cannot provide legally-binding advice. Don’t replace your lawyer. Searching internal legal policies is fine (with citation verification).
Medical diagnosis
Cannot replace a clinician’s diagnosis. Medical use cases need separate review.
Safety-critical autonomous decisions
Hiring, terminations, credit scoring, criminal judgements — anything that should not be auto-decided without human review.
Tasks where accuracy is lower
Excel analysis with complex formulas
Excel analysis with complex formulas
Multi-sheet VLOOKUPs, dynamic-array formulas, complex pivots — LLMs don’t track these well. Where finance accuracy matters, human verification is required.
PDFs with many image-based tables
PDFs with many image-based tables
OCR + table-recognition accuracy is 70–85%. For accuracy-critical accounting data, get the source Excel.
Subtitles for noisy audio recordings
Subtitles for noisy audio recordings
Korean STT degrades on noise, multiple speakers, dialects. Use a good microphone for meeting recordings.
Pinpoint fact retrieval in long documents
Pinpoint fact retrieval in long documents
Finding a single sentence in a 100-page document is sometimes missed due to chunking. Citation page numbers always come back — verify directly.
Truly real-time data
Truly real-time data
Teeem AI pulls data from external systems, but “this exact moment” data depends on sync intervals (typically 1–5 minutes).
Data / security limits
Data residency outside Korea
Data residency outside Korea
Standard deployment is AWS ap-northeast-2 (Seoul). EU/US data residency requires a separate conversation. For multi-region operations, contact sales.
External LLM call (Vertex AI)
External LLM call (Vertex AI)
LLM inference happens on Google Cloud Vertex AI. Data is masked before the call and not used for training (enterprise terms). For “no external service touches our data”, on-prem deployment can be discussed.
Encrypted attachments
Encrypted attachments
Password-protected PDFs / Excels / ZIPs are not auto-processed. The user must unlock or provide an unlocked copy.
Single documents over 500 pages
Single documents over 500 pages
KB chunking applies, but very long single documents may produce inconsistent insights. Split by chapter when possible.
Videos over 500 MB
Videos over 500 MB
Per-file caps are tenant-configured (default 16 MB / request). For large video:
- Upload to external storage (S3, Drive) and pass the URL
- Or split and upload
Channel / integration limits
KakaoTalk — group room limitations
KakaoTalk — group room limitations
Kakao’s official business bot APIs are restricted, so the integration runs through a bridge. Stability is below Slack / Teams; some scenarios (announcement rooms, group-chat manager privileges) are not supported.
Microsoft Teams — some features unsupported
Microsoft Teams — some features unsupported
Adaptive Cards work; Teams Live Events, Walkie Talkie, and a few advanced features don’t integrate.
Browser automation — strong bot detection
Browser automation — strong bot detection
Browserbase stealth mode evades most detection, but CAPTCHA gates can’t be solved automatically. Site terms of service and robots.txt compliance is your responsibility.
Bulk external send
Bulk external send
To prevent spam, mass external email / SMS sends above a threshold trigger a manager approval gate. For marketing campaigns, use dedicated tools.
Model / intelligence limits
No live model training
No live model training
Teeem AI does not train on user conversations (privacy and confidentiality). Company-specific report templates and approval flows must be defined explicitly in the KB or App Pack.
Sentiment / nuance ambiguity
Sentiment / nuance ambiguity
Subtle judgments like “is this message angry?” have variable accuracy in Korean. For things like CS priority auto-tagging, keep human review in the loop.
Long reasoning chains
Long reasoning chains
Multi-step inference of 5–6 hops or more drops in accuracy. Complex decisions need human review.
Creative content (marketing copy, slogans)
Creative content (marketing copy, slogans)
Teeem AI is optimised for company data work. Pure creative copywriting is often better in general-purpose tools like ChatGPT or Claude.
Known issues (subject to change)
This section captures the state at release; some will resolve over time.| Issue | Workaround | ETA |
|---|---|---|
| Very large PPTX (50+ slides) generation time | Split + merge | 2026 Q3 |
| Some KakaoTalk multimedia formats unsupported | Use Slack or web chat | 2026 Q4 |
| Some Microsoft 365 tenant OAuth scope incompatibilities | Admin manually adds manifest | In progress |
What should you use it for?
Pre-rollout self-check: ✅ Strong fit:- Q&A on company documents and policies
- Weekly / monthly report automation
- Account-insight cards
- ERP / groupware data lookups
- Internal tools (inventory / orders / tasks) — App Pack
- Meeting transcription + action items
- Email / quotation / contract draft
- Finance / legal / medical work needing 100% accuracy (human review required)
- Very large data (tens of GB) analysis
- Non-Korea data residency requirements
- Replacing real-time transaction systems
- Autonomous hiring / evaluation / judgement without humans
- Absolute secrets that must not touch any external service (separate discussion required)
Next
Glossary
Definitions for the terminology used here.
Talk to sales
We’ll review your scope together.