# Workflows

> Source: https://whisperer.one/llms/workflows.md · Whisperer machine-readable layer.
> Product overview: https://whisperer.one/llms/product.md · Full corpus: https://whisperer.one/llms-full.txt

How people actually solve tasks with Whisperer, end to end. These are the flows the product
is built around, not illustrations.

## After a meeting: what was said becomes work that is tracked

```
meeting (recorded or imported)
  → live transcript with speakers and timestamps
  → understanding: decisions and commitments extracted, each with its source
  → suggestions offered to the person
  → the person applies them:
        commitment  → task on the board of the meeting's hub, with a due date
        agreement to meet → calendar event
        result worth keeping → knowledge base
  → the hub updates: what is open, what is coming due, what has no owner
  → context persists: months later the answer to "what did we agree" has a source
```

Two properties make this different from an export step. Nothing is applied without the person
confirming it, and everything applied keeps its provenance — the meeting and the moment it
came from.

## Before a meeting: walking in prepared

```
hub for this piece of work
  → previous meetings routed to it
  → open tasks on its board
  → commitments that are still outstanding, and their deadlines
  → files and knowledge attached to the work
  → decisions that were never closed
  → briefing: what is unresolved, not what happened
```

The hub answers "what is missing" rather than "what was said", because before a call the
useful question is what still needs an answer.

## During a call: answering from your own material

```
running transcript
  + knowledge base retrieval (RAG)
  + the user's profile and Leo's memory
  → suggested answer in an overlay invisible to screen sharing
```

The suggestion is composed from the conversation as it runs, not from a single isolated
question — which is why it can answer "what did we quote them last time" rather than only
general knowledge.

## Across meetings: tracing a topic

A question like "how did the pricing decision evolve" is answered by searching across
transcripts, gathering the positions and decisions on that topic, and returning them with
sources — meeting, moment, speaker. The answer is evidence, not a paraphrase.

## From outside: an external AI client working on your Whisperer data

```
Claude / ChatGPT / Cursor
  → OAuth consent, per-domain scopes approved by the user
  → MCP tools: read meetings, search transcripts, create tasks, read files
  → boundary: shared meetings only, nothing irreversible, no acting inside connected services
```

See `/llms/mcp.md` for the exact surface.

## Deferred: asking the assistant to act later

"Remind me in an hour", "send this next week" — stored as a first-class record of what the
user asked for and executed on time, independently of what happened to the system in between.