# Use local AI for synthesis

After core analysis, optional local-LLM modules can summarise and extract commitments. This walkthrough assumes Ollama is already configured.

## Outcome

You will have confirmed Local AI readiness, run (or re-run) summary / [**Meeting extracts**](../runtime/llm.md#modules) modules when available, and spot-checked synthesis against the transcript.

## Starting point

- The planning-review transcript is [imported](../runtime/transcription.md) and a run context is available.
- Local AI is enabled and reachable. Follow [Local LLM (Ollama)](../runtime/llm.md) if you still need setup — this page is not an installer guide.
- You understand that LLM modules send transcript text to your configured Ollama endpoint.

Local AI is **optional**. Workflows 1–3 and 5 work without it.

## What you’ll do

1. Confirm LLM readiness on **Run Analysis**.
2. Include `llm_summary` and meeting-extracts (`llm_action_items`) in the run.
3. Launch analysis and note module status.
4. Read synthesis on Overview / Insights.
5. Check claims against the transcript.

## Walkthrough

1. Open **Run Analysis** for the planning-review transcript. If you only need AI modules on top of an existing core run, choose [**Custom**](../runtime/settings.md#analysis-presets) (or a preset that already includes live-LLM consumers) and ensure transcript summary and meeting-extracts modules are selected. Exact module picker labels follow the registry names surfaced in the UI.

2. Expand **LLM setup**. When Ollama is healthy you should see the active model and optional per-run overrides. If LLM is disabled or unreachable, the panel says so clearly — enable Ollama under [**Settings**](../runtime/settings.md) → Configuration and manage models under **Settings → Models** ([LLM](../runtime/llm.md)), then return here.

![Run Analysis LLM setup section showing that local AI is disabled or not set to Ollama](../_static/workflows/local-ai-llm-setup.png)

3. Choose **Run analysis**. Watch the progress panel: LLM modules may take longer than deterministic ones. Completed, skipped, and failed counts should stay honest if something times out.

4. Open [**Overview**](../public_surfaces.md). After a successful Local AI run, look for summary content and compact meeting extracts with a **Local AI** badge. Until those modules complete, Overview may still show the deterministic summary from core analysis — useful context, but not Local AI output.

![Overview summary area for the planning-review run after analysis](../_static/workflows/local-ai-overview-summary.png)

5. Open **Insights → Actions** for **Meeting extracts** when `llm_action_items` completed. If that module was skipped, open **Insights → Summary** and use the deterministic executive summary as a baseline, then re-run with LLM enabled when you want generated extracts.

![Insights Summary synthesis for the planning review used when checking claims before or beside Local AI extracts](../_static/workflows/local-ai-meeting-extracts.png)

6. Spot-check two claims against **Transcript** (for this fixture: private beta on the twelfth, public launch on the twenty-sixth, analytics deferred). Correct the record in your own notes if synthesis drifts.

7. Treat deterministic modules (highlights, themes, stats) and Local AI synthesis as different kinds of evidence. Prefer the transcript when they conflict.

## What to notice

- **Local AI** marks generated interpretation; it is not the same as deterministic analysis.
- [Balanced](../runtime/settings.md#analysis-presets) may include a limited LLM summary depending on configuration; Custom makes the choice explicit.
- Skipped or failed LLM modules should not wipe the rest of the run.
- Model choice and timeouts live in settings and LLM docs — not in this walkthrough.

## You should now have…

A clear view of Local AI readiness, plus synthesis (Local AI or deterministic baseline) checked against at least two transcript passages.

## Next

- [Export a finished analysis](export-results.md)
- [Local LLM (Ollama)](../runtime/llm.md)
- [Known limitations](../known_limitations.md) for AI behaviour and privacy notes