An AI research assistant that works from your sources
ChatOTL is a research assistant for material you provide: attach papers, reports or transcripts and it extracts findings, compares positions and drafts summaries. It does not browse the web, so every claim traces back to a document you can check.
The dangerous kind of AI research assistant is the one that invents citations. Working from documents you attached removes most of that risk: you can always open the source.
That constraint shapes the workflow below.
A research loop with checkable sources
- Frame the question in the thread — what would count as an answer, what would disprove it.
- Attach sources (PDF, text, Markdown, CSV) and ask for findings, methods and limitations per source.
- Ask where the sources disagree, and why — that's usually the interesting part.
- Draft the synthesis, then verify each claim against the document before you use it.
Reading long documents
Attach up to five files per message, 15 MB each. Text-based formats are extracted into the prompt; PDFs are passed to the model as files. For very long corpora, work source by source in one thread rather than dumping everything at once.
Branch per hypothesis
Fork the thread for each competing explanation. Each branch keeps the shared setup, so you can develop and later compare them without the conversation collapsing into one muddle.
Honest constraints
- No web access: it cannot find literature for you or check what's new.
- It can misread a table or a figure. Verify quantitative claims.
- A summary is not peer review — it compresses, it doesn't validate.
Frequently asked questions
Can it find papers for me?
No. ChatOTL doesn't browse the web. Use a literature database to find sources, then bring them here to read, compare and summarise.
Will it invent citations?
Any model can when asked about things it hasn't been given. Working from attached sources and asking for quotes you can locate in the file keeps this in check.
Is it suitable for unpublished research?
Threads and uploads are scoped to your account. For confidential data, use a self-hosted endpoint so prompts never reach a third party.
How does it help with literature reviews?
One thread per review, one message per source for structured extraction, then a synthesis pass across the thread.
Bring three papers
Extract, compare, then verify. The thread keeps the trail.
Start researching