← All postsBlog

Meet Sophia - the agent that turns meeting transcripts into action items

· Avery NXR

The most underrated productivity loss in modern work isn't bad meetings. It's the time between the meeting ending and the action items being captured, assigned, and communicated. By the time someone gets around to writing the follow-up email, half the room has already forgotten what they committed to.

Sophia is one of the 7 production-ready agent templates that ship with Avery NXR. She watches a folder for meeting transcripts (from Zoom, Otter, Granola, Fireflies, or any other transcription tool you use), extracts the decisions and the owner-tagged action items, and emails each attendee their personalized follow-up.

She runs entirely on your laptop.

What Sophia actually does

Every time a new transcript lands in the folder you configure:

  • Sophia detects the new file
  • Parses the transcript
  • Identifies the participants
  • Extracts every action item with its owner attribution
  • Extracts the decisions made (separate from the actions)
  • Composes a personalized email for each attendee covering only what's theirs
  • Sends each email individually
  • Logs the structured data to wherever you want (a database, a Notion page, an Airtable, a CSV)

The whole loop takes 30-60 seconds per transcript. No transcript or its contents ever crosses to a cloud LLM.

Why personalized matters

The default behavior of most meeting-AI tools is to send everyone the same summary. That's the lowest-effort design but it's also the least useful one. Most attendees skim it and miss what's theirs.

Sophia sends each person only what's theirs — their actions, their commitments, anything they explicitly agreed to. The PM gets the timeline. The engineer gets the technical scope. The salesperson gets the customer follow-up.

Open rates on personalized follow-up emails are 3-5x higher than on group recaps. Completion rates follow.

What's running under the hood

When you open Sophia's graph in the Avery NXR agent builder:

File watcher (transcript folder)
  → Detect new transcript
  → Parse + extract structured content (local LLM)
  → Identify participants (NLP node)
  → Extract action items + owner attribution (local LLM)
  → Extract decisions (local LLM)
  → Loop over participants:
       → Compose personalized email (local LLM)
       → Send via SMTP/Gmail
  → Log structured data to database/Notion/Airtable

Sub-agents handle the per-participant loop. KB retrieval can be added if you want Sophia to reference past meetings or company glossaries while extracting.

What it costs

A cloud-LLM equivalent (sending transcripts through GPT-4 or Claude for extraction + composition) costs roughly $0.20-$0.40 per meeting depending on length. For a team doing 50 meetings a week, that's $500-$1,000 per year per user.

Sophia runs on your local model. Cost per meeting is electricity.

The privacy story

Meetings contain everything: internal strategy, personnel discussions, pricing negotiations, performance reviews, sometimes M&A signal. Most teams don't realize the cumulative sensitivity of their meeting data until they audit the meeting-AI tool they're using.

Sophia eliminates the question entirely. The transcript stays on your laptop. The extraction happens on your laptop. The emails go directly from you to your team. There's no third-party AI provider in the loop.

Try Sophia in 5 minutes

If you've already got Avery NXR:

  1. Open the Agents tab
  2. Find Sophia (SOPHIA · PRODUCTIVITY)
  3. Click "Use this template"
  4. Point her at the folder where your meeting transcripts land
  5. Plug in your email provider
  6. Optionally connect Notion / Airtable / your database for action item logging
  7. Hit Run

Drop a test transcript into the folder. You'll get the personalized emails within a minute.

If you don't have Avery NXR yet, request access at avery.software. Free Desktop tier, no card required.

Why this template, specifically

Meeting follow-up is the single highest-leverage workflow most teams know they should automate and haven't, because:

  • Building it from scratch takes 2-3 days of engineering work
  • Cloud-LLM equivalents have privacy implications most legal teams won't sign off on quickly
  • The off-the-shelf SaaS tools want $20-40/user/month and you still don't own the workflow

Sophia removes all three problems in 5 minutes of setup.