A minimal, adversarial AI assistant that audits and challenges your internal documentation.
Clair is an open-source Python tool designed to test the clarity, consistency, and reliability of your internal knowledge base.
It scans your Notion and Confluence documentation (and soon Slack and others) using LLMs to:
- Detect contradictions between documents
- Flag outdated or unmaintained pages
- Surface vague or ambiguous phrasing
- Identify misalignment between docs and actual team behavior (via Slack integration)
Unlike helpful AI assistants, Clair takes an adversarial approach—seeking flaws, not answers.
- Scan Notion and Confluence pages via official APIs
- Vectorize and compare content to find contradictions
- Detect potentially outdated or unused docs
- Score clarity and usefulness using LLM prompts
- Generate a Markdown report (and optionally write back to Notion, Confluence, or Slack)
- Python 3.10+
notion-clientfor Notion integrationatlassian-python-apifor Confluence supportopenaiorllama-indexfaissorchromadbfor vector similarity- Optional: Slack integration
- CLI (with
typerorclick) - Export to Markdown + Notion + Confluence
- Contradiction detection module
- Outdated/irrelevant page detection
- Slack divergence detection (optional)
- Streamlit UI (optional)
- Full Confluence support: space traversal, content parsing, report writing
Clair is built to challenge—not assist. Its goal is to:
- Stress-test assumptions
- Uncover blind spots in knowledge
- Encourage better writing, clearer thinking, and collective accountability
- Fork the repo
- Create a new branch
- Open a PR with a clear description and sample results
MIT
Cyril Le Mat – Clair was born from a love of clean docs, good questions, and adversarial design thinking.