AI Note-Taking Tool Glossary

Terms a professional encounters when choosing an AI note-taking tool, simply defined. Each definition fits in one to three sentences, written to be used verbatim by search engines or AI. For purchasing questions, the compliance and sovereignty FAQ answers point by point; to see these components working together, try Leexi in one of your meetings.

Product and Summary

AI note-taking tool. Software that joins your meeting (video, phone, or in-person), transcribes it, and drafts the summary: decisions, actions, deadlines. It writes notes for you instead of helping you write your own. Full details at AI meeting note-taking.

Notetaker. Common English term for an AI note-taking tool. At Leexi, we say “AI note-taking tool.”

Summary (CR). The document written after the meeting: recap, decisions made, assigned tasks, and deadlines. It’s the deliverable, unlike the raw transcription. Some meeting summary tools produce it automatically.

Minutes (PV). Formal summary of a body: board, general assembly, works council, municipal council. Same logic as a summary but with stricter format requirements.

Transcription. Word-for-word text of what was said during the meeting. It serves as the basis for the summary but remains raw and must be processed to extract essentials.

Speaker identification (who speaks). Assigning each part of the transcription to the person who said it. This allows knowing who made which commitment.

Automatic summary. Synthesis generated from the transcription: the meeting’s essentials without reading it all.

Summary template. A summary model adapted to a meeting type, e.g., recruitment interview or sales meeting. The summary then highlights what matters for that meeting type.

Library. History of your recorded meetings, accessible. You can find past conversations without reopening recordings.

The assistant (the bot). The agent joining the meeting to record and transcribe it. It is visible in the meeting, as required by the EU AI Act.

Autofill CRM. Automatic filling of CRM fields from the summary without re-entry. The right summary links to the right record via CRM and video integrations.

In-person. A meeting in a room, as opposed to video. An AI note-taking tool working only in video misses these; the mobile app records the room.

Compliance and Sovereignty

GDPR. The European personal data regulation. For an AI note-taking tool, it requires a legal basis for recording, participant information, and controlled retention.

Legal basis. The legal ground authorizing recording. Internally, employers rely on legitimate interest after balancing tests and prior notice; for calls with externals, explicit consent is usually required.

Timestamped consent. Dated proof that participants were informed and consented to recording, kept in an exportable log configurable by team and country.

ISO 27001. A voluntary information security standard. Not legally required but a de facto prerequisite in B2B bids, large accounts, and public contracts. Details on certifications, audits, and compliance.

ISO 42001. AI system management standard. At Leexi, it is in progress.

NIS2. European directive imposing legal security obligations on “essential” and “important” entities. It creates the obligation, not ISO 27001, which is a means to comply.

EU AI Act. European AI framework. It mandates transparency (visible assistant in meetings) and impact analyses for sensitive uses. Leexi’s commitment on the EU AI Act is documented.

Cloud Act. US law allowing authorities to demand access to data held by US companies, even if stored elsewhere. Hence, AWS hosting in France alone doesn’t meet strict sovereignty.

Sovereign hosting. Hosting meeting three conditions: datacenters physically in France, legal entity not subject to extraterritorial laws, and European ownership control.

HDS (Health Data Host). Mandatory certification to host personal health data in France. Without it, a tool cannot process such data for health organizations.

SecNumCloud. ANSSI qualification certifying high security and immunity to extra-European laws. Mainly concerns public sector sensitive data.

DPA (data processing agreement). Contract governing your data processing by a provider. One of the first documents a DPO requests before purchase.

Subprocessor. A provider processing data on behalf of the supplier. The subprocessor list is part of evidence buyers require to map the processing chain.

No training. Your conversations are not used to train AI models. This is the first check for technical teams.

Closed loop. Processing where the language model runs in a controlled perimeter, not exposing your data to third-party use. At Leexi, the model runs via Azure France in a closed loop.

Encryption in transit and at rest. Protecting data during transfer and storage.

Schrems II / Data Privacy Framework. Legal framework governing EU-to-US personal data transfers. Tools hosting and processing entirely in the US depend on it.

SOC 2. US-origin audit report, functional equivalent of ISO 27001. Buyers request one or the other as third-party proof, not just a declaration.

DPIA / AIPD (impact analysis). Data protection impact analysis required for sensitive processing. The EU AI Act makes it almost systematic before deploying AI tools.

Pseudonymization and anonymization. Pseudonymization replaces identifiers with reversible codes; anonymization makes identification impossible. Only anonymization excludes data from GDPR.

Personally Identifiable Information (PII). Any info identifying a person directly or indirectly. Meeting transcriptions almost always contain some: names, roles, spoken content.

Data residency. Where your data is physically stored. For Leexi, it is France.

Under the Hood: Technology

Speech recognition (ASR, speech-to-text). Converting speech to text. The first step: without reliable transcription, no reliable summary.

Diarization. Technical term for speaker identification: segmenting audio and assigning each part to the right person. Enables writing “who said what.”

Voice activity detection (VAD). Detecting when someone speaks to transcribe speech and ignore silence and background noise.

Word error rate (WER). Quality measure of transcription: percentage of mis-transcribed words. Lower is better; often French and Dutch distinguish tools.

Natural language processing (NLP). Techniques enabling machines to analyze and produce language. Transforms raw transcription into structured summary.

Large language model (LLM). AI model drafting summaries, decisions, and tasks from transcription. At Leexi, an OpenAI model runs via Azure France in a closed loop.

Prompt and prompt template. Instructions given to the model to produce the summary. One template per meeting type guides the summary to what matters for that exchange.

RAG (retrieval-augmented generation). Technique where the model answers using specific documents rather than only memory. At Leexi, “Ask Leexi” queries your meeting history.

Extractive and abstractive summary. Extractive selects exact sentences; abstractive rephrases. Good summaries combine both: reformulated decisions, exact quotes when authoritative.

Hallucination. When a model produces plausible but false info. A risk to monitor in summaries, and why nothing is sent before your review.

Domain vocabulary / custom dictionary. Adding sector-specific terms (product names, acronyms, jargon) so transcription spells them correctly.

Latency and real-time. Delay between speech and transcription. Real-time shows transcription during the meeting, not after.

API and webhook. Technical connection points sending summaries automatically to other tools (CRM, knowledge base) without human action.

TLS and AES encryption. TLS protects data in transfer, AES during storage. This is the concrete meaning of “encrypted in transit and at rest.”

Integrations and Deployment

CRM. Tool centralizing contacts and opportunities (HubSpot, Salesforce, Pipedrive, Dynamics). The goal for an AI note-taking tool is to send summaries to the right place.

VoIP. Internet telephony (Aircall, Ringover). Enables capturing and summarizing calls, not just video meetings.

SSO / SAML. Single sign-on connecting the tool to the company directory. Setup often done by IT admin.

Calendar connection. Linking the tool to your calendar (Outlook, Google Calendar), allowing the assistant to join scheduled meetings automatically.