ProvenanceLog
A local-first app that records which AI agent session, prompt, or manual edit produced each figure, table, or draft in your thesis folder — built so graduate students can paste clean methods-section copy.
graduate students and postdocs whose figures and drafts increasingly come from AI coding agents
- Watch any thesis folder and link each file to the AI session, prompt, or human edit that produced it
- One-click methods-section excerpt generating 'file X was produced by session Y on date Z'
- Local-only storage so no data leaves the laptop — IRB and journal-friendly by default
- Notebook-aware tracking across Jupyter, Quarto, and Word outputs
AI-disclosure requirements are tightening across journals and academic integrity policies just as coding agents become standard in research workflows — reproducibility gaps are now a personal liability.
Multiple recent sources confirm tightening AI disclosure rules across journals and university policies (Tandfonline 2025 articles, ICMJE, gradschoolcenter.com), but the specific use case — tracking per-figure/table provenance for thesis writing — is a narrower slice of a broader trend.Full article: Disclosing artificial intelligence use in scientific ↗AI and Academic Publishing Ethics for Graduate Students ↗
Direct overlap from multiple local-first AI session/attribution tools already shipping: CodeBurn (free, tracks 36 AI coding tools), AgentDiff (AI code attribution), tokenlens, and Agent Beacon — though they primarily target code files rather than figures/tables/academic drafts, the technical approach is the same.GitHub - getagentseal/codeburn: Free, local tool to track AI coding ↗AI Code Attribution: What It Is and How to Track It — AgentDiff ↗
Target market is graduate students and postdocs — historically price-sensitive; dominant competitors (CodeBurn, AgentDiff, tokenlens, Agent Beacon) are free or open-source, giving users no reason to pay for a thesis-specific wrapper.GitHub - getagentseal/codeburn: Free, local tool to track AI coding ↗GitHub - ATOM00blue/tokenlens: Where are your AI agent's tokens going ↗
AI disclosure mandates are actively expanding across publishers (Nature, Elsevier, ICMJE) and universities, and journal review processes now routinely flag undisclosed AI use — this pressure is structural and unlikely to reverse.AI Policies in Academic Publishing 2025: Guide & Checklist ↗Full article: When and how to disclose AI use in academic publishing ↗
Proven technical pattern — multiple existing tools already read local session logs from Claude Code, Cursor, Codex and stitch per-file attribution; extending from code files to figures/drafts is incremental rather than novel.GitHub - getagentseal/codeburn: Free, local tool to track AI coding ↗PROV-AGENT: Unified Provenance for Tracking AI Agent Interactions ↗