Glossary of key entities

Overview

This glossary collects definitions of Whisperer's core concepts — from the call session to the knowledge base and plans. It exists so the rest of the documentation reads without guesswork: when you encounter a term in any article, you'll find its exact meaning here.

This is a reference page. Instead of a step-by-step procedure, it includes a "How the entities connect" section — it shows how the concepts come together into a single product workflow.

When to use

You're reading the documentation for the first time and want to check the terms.
Another article used an unfamiliar term (for example, "no-logs" or "model role").
You're configuring prompts, the knowledge base, or a subscription and want to understand the boundaries of the concepts.

Terms

Session — one assistant session on a call. It has a lifecycle: start (session.start) → audio recording → questions and answers → finish with data saved. A session has a transcript, the questions asked, and the model's answers attached to it. The session is the unit of minute accounting and the basis of history in the dashboard.

Transcript segment (Segment) — a fragment of recognized speech that Whisperer forms on the fly from streaming audio. Each segment is tagged with a speaker and a timestamp. Segments add up to the session's full transcript, shown as a ticker in the client and saved in history (except in no-logs mode).

Speaker ([Me] / [Other]) — the speaker label in the transcript. [Me] is your voice from the microphone, [Other] is the voice from system audio (what plays in your headphones/speakers). This labeling is possible because the client captures two independent audio sources and tags each audio chunk with a speaker label.

Mind-map (node: topic / action_item / decision) — the meeting map that Whisperer extracts after a session. Nodes (MeetingMapNode) come in three types: topic (a discussed topic), action_item (a task to be done), decision (a decision made). In the dashboard the map can be visualized and further enriched (enrich-mindmap).

Prompt template (role prompt) — a draft system instruction for the model tailored to a specific role. Set by you in Prompt Studio and embedded into the system prompt before the user context and the knowledge-base block.

Model role — the purpose a model is selected for. Whisperer has seven roles: responses (regular chat answers), coding (code), vision (screenshots and images), system_design (structured response with Mermaid diagrams), generation (generating and enriching the mind-map), transcription (speech recognition, Whisper), embedding (text vectorization for RAG). You can set a separate model for each role.

Knowledge base / Note (Note) — your personal knowledge store in the dashboard. A note contains a title, Markdown text, and tags. Wiki-links of the form [[...]] connect notes into a graph (neighbors, shortest path, centrality). The knowledge base is kept until you delete it.

RAG (Retrieval-Augmented Generation) — the mechanism by which the model's answer is augmented with facts from your knowledge base. Notes are split into chunks (~512 tokens, overlap 64), turned into embeddings, and stored in a vector database; for each query the system uses cosine search to find the top-k relevant fragments and inserts them into the prompt. This way the suggestions rely on your materials, not just the model's general knowledge.

User context — arbitrary data you provide to the assistant in advance (for example, a description of your role, product, or résumé). Set in your profile, it influences the tone and depth of responses across all sessions.

Subscription / Plan (Plan) — your level of access to the product: Free, Start, Pro, Max. The plan determines the minute limit, access to the knowledge base, multi-screenshot, System Design mode, and frontier models. The product's core (sessions, knowledge base) requires an active paid subscription.

Minute quota — the recognition-time limit within a plan. On Free it's 60 minutes per month; on paid plans minutes are unlimited. A session's minutes count toward the quota, including in no-logs mode (a separate counter, no_logs_minutes_used). The quota is soft: the current session will finish, but you won't be able to start a new one until it's replenished.

No-logs (ephemeral) mode — a private session mode in which the transcript and answers are not written to the database and are deleted after the session ends. Useful for sensitive conversations. Important: minutes are still spent from the quota.

Frontier models — flagship, most powerful LLMs (for example, Claude Opus, top-tier GPT, and equivalents). In Whisperer they're unlocked only on Max plans.

Referral — a user you invited. For each qualifying purchase by a referral (Pro or Max plan) you get a 500 ₽ reward; one reward per invitee.

Promo code — a discount code. Two types: percent (a percentage discount off the price) and bonus_days (bonus subscription days). The currency-conversion markup is not applied to discounts and bonuses.

How the entities connect

Whisperer's concepts form a single flow — from conversation to knowledge:

You start a session. The client captures two audio sources and tags them with speakers [Me] / [Other].
Recognition (the transcription model role) turns audio into transcript segments.
When you ask a question, the system prompt is assembled: the chosen model role's prompt template → user context → the RAG block from your knowledge base.
The model answers; in no-logs mode neither the transcript nor the answers are saved, but minutes are still deducted from the quota set by the plan.
After the session, Whisperer extracts a mind-map (topic / action_item / decision) for the meeting summary.
The level of access to all of this is set by the subscription; promo codes and referrals help extend or cheapen it, and frontier models are enabled on Max.

Screenshots

📸 [Screenshot: live transcript with [Me] and [Other] labels]

📸 [Screenshot: meeting mind-map with topic/action_item/decision nodes]

📸 [Screenshot: knowledge-base graph built from wiki-links]

Common mistakes

Confusing a session with a transcript. A session is the whole session, with minute accounting; the transcript (a set of segments) is only part of it.
Treating no-logs as "free." Minutes in no-logs are spent as usual; only the data isn't saved.
Expecting frontier models on Pro. Frontier is available only on Max.
Trying to issue commands through user context. Context provides facts about you; use a Prompt Studio template to control how the assistant responds.

Best practices

Check the glossary before configuring prompts and model roles — it saves debugging time.
Fill in your user context and knowledge base in advance: RAG pulls your facts into every answer.
Use wiki-links between notes so the knowledge graph helps with search and navigation.
For sensitive calls, enable no-logs, keeping in mind that minutes are still spent.