How the AI works
Reading a meeting is a language problem
People do not speak in decisions and action items. They interrupt, backtrack, and imply. Pulling the commitments out of that, with the right owner and the right date, is language work, and only recent models are reliable enough to do it well. The model is the engine of Sol, not a feature bolted onto a notes app.
From raw transcript to owned actions
Read the transcript
Sol takes the full meeting transcript, however long and however informal, and prepares it for the model. No structure is assumed. People interrupt, backtrack, and leave things implied, and Sol reads all of it.
Extract decisions and owned actions
A language model reads the whole conversation and identifies the choices that were actually made, not just discussed, and the commitments each person took on. Owner and due date come only from what was said. If neither was stated, Sol marks the item accordingly rather than inventing either.
Write a short summary
A second, faster model writes two to four sentences a person who missed the meeting can read in under a minute. Context for the decisions, not a transcript of the discussion.
Track completion across meetings
At each subsequent meeting, Sol matches the new transcript against open items from earlier meetings and marks what got done, what is still open, and what is overdue. The record carries forward automatically.
Each step goes to the right model
Extracting decisions and tracking completions are very different tasks and do not need the same horsepower. Sol keeps a registry of frontier and open models and routes each step to the one that clears the quality bar for the least cost. Relying on a single lab is a reliability risk of its own.
Read the whole transcript and pull out the decisions that were made and the action items, each with an owner and a due date, including commitments that were only implied. The heaviest step, sent to a frontier model.
Write a short, plain-language summary that a person who missed the meeting can read in under a minute. Cheaper and quick.
Match this meeting's actions against open items from earlier meetings and mark what got done, what is still open, and what is overdue.
long transcripts · implicit commitments · who owns what
structured extraction · summaries · fast turnaround
short summaries · matching actions across meetings
careful reading · no invented owners or dates
self-host · data residency · cost floor
Kept honest and kept private
Grounded in what was said
Every owner and every due date Sol assigns comes from the words spoken in the transcript. If the meeting did not name a person or a date for an action, Sol marks it unassigned or without a date. It never fills a gap or invents a commitment. A wrong read is quick to spot because the output is short and specific.
Your transcripts produce your outcomes only
Meeting transcripts are used to extract decisions and actions for your workspace and are never used to train a shared model. Encryption in transit and at rest, retention controls, and deletion on request. On the Business plan, route to open models you host yourself so nothing leaves your perimeter. Read the Security page.
A note on this deployment: the OpenAI adapter is wired live and the router sends each step to the appropriate OpenAI-hosted model. The other providers in the registry share the same interface and activate once a key for them is configured. The routing table above is the real table the engine uses, not a simplified version of it.
See it read your own meeting.
Start free, no card, and turn your next transcript into decisions and owned actions in about a minute. Or book a walkthrough for your team.