An open-source fact-checking pipeline for news articles, Frank Investigator aims to combat bias and misinformation. It targets engineers and researchers needing to audit journalism, providing structured verdicts through an evidence graph, source authority analysis, and multi-model LLM consensus, prioritizing factual accuracy over popular consensus.

How It Works

The system fetches articles via headless Chromium, extracts canonical claims, and builds an evidence graph by following in-body citations. It employs "Authority trumps volume," primary source veto, and confidence-weighted LLM consensus. Key features include defense against smear campaigns, detection of 16 rhetorical fallacies, and penalization of echo chambers and headline-baiting.

Quick Start & Requirements

Setup involves bin/setup for dependencies and bin/dev to start the server. Essential requirements include LLM API keys (OPENROUTER_API_KEY or OPENAI_API_KEY) and configuration for the LLM provider and models. Dependencies include SQLite3, Solid Queue, Solid Cable, Solid Cache, and Headless Chromium. The UI is server-rendered.

Highlighted Details

• Extracts claims, follows citations, and constructs an evidence graph.
• Classifies source authority and role (e.g., government, reporting, opinion).
• Detects 16 rhetorical fallacies and various manipulation tactics (temporal, statistical, quotation, authority laundering, contextual gaps).
• Analyzes headline-body divergence, coordinated narrative campaigns, and emotional manipulation.
• Offers live updates via Turbo Streams and hybrid retrieval for related investigations.

Maintenance & Community

Deployment is managed via Kamal, with commands for re-analyzing, refreshing, and cross-referencing investigations. The README does not list specific community links (Discord, Slack) or notable contributors.

Licensing & Compatibility

Licensed under the GNU Affero General Public License v3.0 (AGPL-3.0). This strong copyleft license requires derivative works distributed over a network to be made available under the same terms, potentially restricting integration into closed-source commercial products.

Limitations & Caveats

Functionality relies on external LLM API keys. The AGPL-3.0 license imposes significant obligations on distributed use. While supporting international sources, the project emphasizes Brazil-first source profiling.