Skip to main content
AutoCleanEEG can optionally generate textual summaries using a large language model. These reports live under the task‑root reports/llm/ folder and are built from the processing log and PDF report.
By default AI reporting is disabled. Set "ai_reporting": true in task config and provide OPENAI_API_KEY to enable.
Privacy Considerations for LLM ReportingFor 100% privacy, use a local language model instead of cloud-based APIs. When using OpenAI’s API, ensure your setup includes zero data retention and a Business Associate Agreement (BAA) if handling Protected Health Information (PHI). Note that most data sent is deidentified and consists of aggregate processing metrics, not raw EEG data.

​
Files Produced

FilePurpose
context.jsonSerialized run context used to produce all text.
methods.mdDeterministic methods paragraph created without any API calls.
executive_summary.mdStudy-ready summary produced by the LLM (requires API key).
qc_narrative.mdLLM-generated quality control narrative and recommendations (requires API key).
llm_trace.jsonlHash-based trace of prompts and results for compliance.

​
Enabling the Feature

  1. Add "ai_reporting": true to your task configuration or workspace template.
  2. Ensure an OPENAI_API_KEY is available in the environment.
  3. Run the pipeline as usual – reports appear under reports/llm/ (or a subfolder keyed by the subject base name).

​
CLI Usage

You can regenerate reports or chat about a run from the command line: 
autocleaneeg-pipeline report create --run-id demo --context-json ./context.json \
  --out-dir ./reports
autocleaneeg-pipeline report chat --context-json ./context.json
report create always writes context.json and methods.md. If an API key is present, it also generates executive_summary.md and qc_narrative.md.

​
When to Use

  • Share short summaries with collaborators.
  • Capture deterministic methods text for manuscripts.
  • Quickly review quality metrics without opening the full PDF.
Missing API keys or expected input files never break your run; the pipeline simply skips LLM outputs.