Notes

Tracking founder 1:1s without a CRM

Updated April 2026

A founder's hardest 1:1 problem isn't taking notes — it's recalling, six weeks later, what you and a specific person decided. Gistlist records each 1:1 on your Mac, transcribes it locally, summarizes it with a prompt you choose (local Ollama or your own Claude / OpenAI key), and lets you query the whole library through Claude Desktop. No CRM, no shared notes service, no hosted Gistlist account.

The actual problem

A founder running weekly 1:1s with five direct reports has roughly 250 conversations a year. Each one contains commitments — “I’ll get back to you on the hiring loop next week”, “let’s revisit the pricing question after the board meeting” — that need to be honored without re-reading every transcript.

The problem isn’t note-taking during the meeting. The problem is recall, six weeks later, when “didn’t we already talk about this?” is the right question and you can’t quite remember.

Existing options each fail in their own way:

  • A notebook: ground truth, but unsearchable across months.
  • A shared notes app: searchable, but you write differently when the doc is shared. The candid version stays in your head.
  • A CRM: built for sales, not for relationships you don’t want to instrument.
  • An AI meeting tool that uploads everything: searchable and candid, but now your private 1:1s live on someone else’s server.

The ideal: every conversation captured verbatim, summarized into something scannable, fully searchable — with files that stay on your disk and a configuration you control over which cloud providers (if any) see the contents.

The workflow

Gistlist breaks the loop into four steps you only have to set up once.

1. Record. Open Gistlist, hit record, do the meeting. Mic capture uses a native CoreAudio helper for clean USB-mic input, with an FFmpeg AVFoundation fallback when the helper isn’t available. System audio uses AudioTee, a CoreAudio tap, on macOS 14.2 and later. Pause and resume mid-meeting are first-class operations.

2. Transcribe. On Apple Silicon, transcription runs locally through Parakeet — Nvidia’s open-weight ASR model, executed via MLX on your machine. No audio leaves your laptop on the Parakeet path. If you’d rather pay per minute and skip the local install, the Setup Wizard also exposes OpenAI Whisper (cloud).

3. Summarize with a prompt you choose. The default Setup Wizard configuration runs summaries on local Ollama, fully on your machine — Qwen by default, or any other model you’ve pulled (Llama, Mistral, etc.). If you’d rather use a cloud model, switch the LLM provider to Anthropic Claude or OpenAI in Settings; the call goes directly from your machine to that provider with your API key. Either way, the prompt is yours to edit; the default captures a meeting at a glance, and for 1:1s specifically you’ll want a custom one (see below) that pulls out commitments, recurring themes, and follow-ups.

4. Query through Claude. Connect Gistlist’s MCP server to Claude Desktop (one config file edit, details here) and Claude can answer questions across your entire meeting library on demand.

A custom 1:1 prompt

Prompts live as plain markdown files with frontmatter — see the prompts docs for the spec. A workable starting point for 1:1s:

---
title: 1:1 review
description: Pull out commitments, recurring themes, and follow-ups from a 1:1.
default_for: meeting
---
You are reviewing a transcript of a recurring 1:1 conversation.
Produce a brief markdown report with these sections:
- **What I committed to** — anything I said I'd do, with rough timing if mentioned.
- **What they committed to** — same, from their side.
- **Open questions** — anything either of us flagged as unresolved.
- **Themes** — recurring topics or tensions worth tracking across meetings.
- **Verbatim quotes worth keeping** — short pulls that capture how something
was actually said.
Be concrete. Use bullet points. Keep it under 300 words.

Gistlist runs this prompt automatically when the meeting ends, writing the output as a markdown file alongside the transcript.

Asking Claude across your meeting library

This is where the system pays off. With the MCP server connected, Claude can search and read across every meeting you’ve recorded. Some queries that work well:

“What did Jordan and I decide about the launch timeline last quarter?”

Claude calls search_meetings for “Jordan launch timeline”, finds the relevant runs, calls get_meeting on each, and summarizes the decision and how it evolved.

“What’s the most recent thing I committed to following up on with Maya?”

Claude lists Maya’s recent meetings, pulls the most recent summary, and surfaces the open commitment from your “What I committed to” section.

“Has anyone mentioned concerns about the new performance review process more than once?”

A pattern query: Claude searches across all 1:1s, identifies the people and instances, and tells you which themes are recurring instead of one-off.

The conversations stay on your disk; the answers come out of Claude. That separation is the whole point.

Why local-first matters here

For most meeting tools, “local-first” reads as a privacy talking point. For 1:1 tracking specifically, it’s the load-bearing requirement.

A 1:1 is the meeting where someone tells you they’re considering leaving, where you say something honest about a peer, where a hiring decision actually gets made. None of that should sit in a hosted notes service waiting to become a court exhibit, a leak, or training data. Keeping it on your disk makes the candid version writable.

The downside — no team sharing, no cross-org visibility — is the right downside for this kind of meeting. A 1:1 is supposed to be a closed loop between two people; the tool should respect that, not unlock it.

For meetings that should be shared (customer calls, all-hands), Granola vs Gistlist walks through where each tool is the right call.

Further reading

Frequently asked questions

Can Claude really read across all my meetings?

Yes, through Gistlist's local MCP server. The server exposes three tools to Claude Desktop — list_recent_meetings, search_meetings, and get_meeting — that scan a local SQLite index of your meeting library. Claude calls them on demand to answer questions like "what did Jordan and I decide about the launch timeline last quarter?"

Where are my meeting files actually stored?

Locally, by default in ~/Documents/Gistlist/ — one folder per meeting containing the audio, transcript, summary, and any prompt outputs. You can point Gistlist at a different folder during setup, including a folder inside an Obsidian vault. Gistlist does not upload these files to a Gistlist-hosted service; the only outbound calls are to whichever cloud provider you've selected for transcription or summaries (Claude, OpenAI, OpenAI Whisper), and only for the data those calls require.

Can I customize the summary?

Yes. Gistlist ships one default summary prompt and lets you add as many of your own as you want — a 1:1 review prompt, a decision log, a follow-up email draft, a customer recap. Prompts are plain markdown files with frontmatter; see the prompts docs for the format.

Can I pause and resume a recording mid-meeting?

Yes. Audio is written to disk continuously rather than buffered in memory, and pause/resume during a meeting is a first-class operation — handy when a call gets interrupted or you want to drop a stretch of small talk.

Do I need an Anthropic or OpenAI API key?

It depends on the providers you pick. The default Setup Wizard configuration uses Anthropic Claude for summaries (your API key, your billing relationship). For a key-free path, switch the LLM provider to local Ollama in Settings — combined with Parakeet for transcription on Apple Silicon, the whole recording → summary pipeline then runs on your machine. Querying through Claude Desktop is a separate piece: it requires Claude Desktop, which is a consumer Anthropic product you sign in to.

Why not just use a CRM for this?

Two reasons. CRMs assume a customer-facing relationship and are awkward for internal 1:1s. And the moment you put your candid notes about a teammate into a shared system, you write more carefully — which means you write less of what you actually need to remember. A local recording lets you be honest with yourself.

← All notes