kayetanaResearch with AI before you publish.
New results, lab protocols and invention disclosures are often the most valuable things a lab has. kayetana helps you read, search and summarize, without any of it leaving your network.
Query the literature, with a source for every sentence.
Load papers, reports and lab protocols as PDF or Word files and ask your question. kayetana answers only from these texts and numbers its sources. That way, you can check every statement against the original.
- Several documents in one request
- A source list under every answer
- Also via API with citations
Where kayetana helps in the lab and in development.
Internal patent search
Search your own collection of patents and invention disclosures with embed and rerank.
Review literature
Summarize methods, findings and open questions across many papers.
Compare experiments
Put two test series side by side and pull out the differences.
Capture meetings
Record lab meetings and file them as text with timestamps.
Embed entire collections
Embed datasets in the background and reuse the result as JSONL.
Organize documents
Classify reports by project, method or status.
Models you choose yourself.
kayetana ships with openly licensed models for chat, search and speech. You can pull more open models by name. Once they're loaded, everything runs without internet, even on isolated networks.
- Apache 2.0 or MIT licenses
- A default model per task
- Size and context length at a glance
How your team gets going.
Choose a collection
For example, the literature for one project or your own invention disclosures.
Upload
Load documents as a dataset or use them directly in the playground.
Search and ask
Find with embed and rerank, then dig in with answers from documents.
Grant access
Invite colleagues by link and assign roles.
See kayetana with sample literature.
In the demo, we query research papers with citations and search a sample collection.
Frequently asked questions
Do unpublished results stay in-house?
Yes. kayetana runs on your own server. Prompts and responses aren't stored, and once the models are loaded, no internet is needed.
Does kayetana search external patent databases?
No. kayetana only searches the documents you upload yourself, such as your own collection of patents.
How does search across many documents work?
Embed turns texts into vectors, and rerank sorts them by similarity to your question. Both run on bge-m3.
Can I use larger models?
Yes, with more memory in the Mac Studio. We recommend 36 GB or more so models up to about 20 billion parameters run alongside search and transcription.