> For the complete documentation index, see [llms.txt](https://oodlz-ai-studio.gitbook.io/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://oodlz-ai-studio.gitbook.io/docs/reporting-agents/growth/monthly-summaries.md).

# Monthly Summaries

LLM-generated 3-bullet plain-English summary per channel per month — temperature 0.3, fixed prompt, input snapshot stored alongside for traceability.

> **3 plain-English bullets per channel — facts and direction, no jargon, no invention.**

Monthly Summaries is the LLM layer over the metric data. After each channel's workflow writes its monthly metrics, a sub-workflow runs the snapshot through an LLM and produces 3 plain-English bullets summarising the period — facts and direction only. The summary lands in the workspace next to the metric record and renders in the dashboard's AI Summary card.

## How It Works

Every channel's workflow calls the Monthly Summaries sub-workflow at the end of its run, passing the channel name, period label, and the metric snapshot (including the prior-period values). The sub-workflow assembles a fixed prompt — describe what happened using these numbers, no recommendations, no invented context — and calls the LLM at temperature 0.3 for low variance.

The LLM returns 3 bullets, each under 280 characters, each starting with a clear fact and a direction (improved by X%, declined by Y%, flat vs prior month). The output is stripped of any markdown the LLM might have added, written to the workspace, and tagged with the model used and the input snapshot.

The input snapshot — the exact metric numbers fed to the LLM — is stored alongside the summary text. If the summary ever needs to be regenerated, the input is reproducible without re-fetching from the platform APIs.

## What You Get

* 3 plain-English bullets per channel per month — covering the period's headline facts and direction.
* Low-variance generation — temperature 0.3 keeps tone factual and consistent across runs.
* No recommendations, no invented context — the prompt explicitly excludes both.
* Input snapshot stored with the summary — every bullet traces back to the exact numbers it described.
* Model used logged on every summary — audit trail for any regenerations.