Meeting summaries written on your Mac.

The useful parts of a call, in one place

After a meeting, interview, or long discussion, a full transcript is useful for checking the details. A summary helps you find the main points first. echo99 uses a local language model to turn a transcribed recording into readable notes.

The default instructions ask for an overview, key points, decisions, action items, and open questions, in the language spoken in the transcript. Action items include an owner when the conversation names one. You can edit those instructions in Config → LLM to suit your own notes.

How to summarize a recording

1. Open Config → LLM, choose a model for summarization, and install it. The default, Gemma 4 E2B, is a one-time download of about 2.8 GB.
2. Record a call or import an existing recording, then let echo99 finish its transcription.
3. Open the recording in Recordings, select Summary, and generate the summary.
4. For future recordings, enable Summarize automatically after transcription in Config → LLM to add a summary after the transcript is ready.

You can also summarize older recordings that already have a transcript. If the transcript changes after the summary was written, for example because you named a speaker or transcribed the recording again, the Summary pane says so, and Summarize Again brings it up to date. Use it too when you want to try different instructions.

Your notes stay with your recording

The model runs entirely on your Mac. After its initial download, summarization works offline: your transcript is never uploaded for processing, and there is no account to create.

Read the result in the Summary pane or use summary.md, the Markdown file saved in the recording folder. Your audio and full transcript stay available, so you can return to the original wording whenever a detail matters.

What to expect

• A summary depends on both the transcript and the model. Check names, decisions, and commitments against the transcript or audio before relying on them.
• Models need disk space and memory. The four choices are a one-time download of about 1.3 to 7.2 GB and need about 2.5 to 12 GB of memory while they run; the default, Gemma 4 E2B, needs about 4.5 GB. The technology page compares them.
• On Apple silicon the model runs on the GPU through Metal. On an Intel Mac it runs on the CPU, which is slower. Either way a summary can take several minutes.
• By default echo99 loads the model when it has a summary to write and frees the memory once the model sits idle. Turn on Keep model loaded in Config → LLM to keep it in memory between summaries.
• The instructions and transcript together must fit in about 30,000 tokens, roughly 20,000 words of English (fewer in Czech). A longer transcript is not summarized at all: echo99 reports “Transcript too long for summarization” instead of silently summarizing only part of the conversation.

New to echo99? Follow the getting started guide to make your first recording.

Related guides: Local transcription, Live transcription, Meeting notes without a bot