Summaries

A summary is written by a language model and saved as a report.

ailoud audio summarize ID001

There is no default selection. Name the recordings, or a tag:

ailoud audio summarize ID001 ID002
ailoud audio summarize --tag standup

Shape it

ailoud audio summarize ID001 --template one-on-one

The headings change with the template. See Templates.

Tell it what the transcript does not say

ailoud audio summarize ID001 \
  --template one-on-one \
  --context "Ann is Ben's manager; this is their fortnightly."

Keep it to a sentence or two. Every word competes with the transcript for the model's attention.

Choose the language

ailoud audio summarize ID001 --lang en

Without --lang, the summary is written in the language of the recording.

Groups

Several recordings become one report, not several stapled together:

ailoud audio summarize --tag offsite

A group reuses each recording's stored report instead of re-reading its transcript, which is much cheaper. Force a fresh read:

ailoud audio summarize --tag offsite --fresh

A single recording is always read from its transcript. A summary of a summary drifts.

Reports

ailoud report ls
ailoud report ls --recording ID001
ailoud report show SUM0
ailoud report rm SUM0

SUM0AB...  2026.08.31 12:54  haiku  ru  01M1B2W5EG  Встреча у пирса в 5:00

Each report records the template it used, the context it was given and the model that wrote it.

Deleting a report leaves the recording and its transcript alone. You can make it again.

Which model

Set llm.provider in the config file:

Provider	Reaches	Credential
llama-cpp	a local GGUF model	none
claude-cli	Claude via Claude Code	its own sign-in
anthropic	Claude API	ANTHROPIC_API_KEY
openai-compatible	OpenAI, Ollama, LM Studio	OPENAI_API_KEY

ailoud setup asks and writes the answer for you.

Long recordings

A transcript too long for the model is summarised in portions and the answers combined. You see the progress:

Summarising portion 3/8 (37%)