Two AI products,
live today.
Evi is a personal AI assistant that remembers you. Kus makes small models answer like big ones. Both designed, built, and run by us — on infrastructure we operate ourselves.
Most assistants start typing. Evi plans first.
It works out what kind of question you asked and decides how to answer before writing a word — then marks its own work.
- It plans the answer before writing it
- The question is classified, then a separate pass picks the format, length and tone and lists what to cover. A debugging question leads with code; a maths question gets worked through; 'hi' gets one line.
- It marks its own work
- Every reply is scored out of 1 after streaming — against the plan, and against the search results it was actually given rather than its own training.
- Memory and documents come as standard
- Long-term memory across chats and answers drawn from the PDFs you upload, tied back to the source text.
Make a small model answer like a big one.
One API that searches trusted sources and answers with citations — turning weak or local LLMs into reliable ones.
- Hierarchical parallel retrieval
- A master router fans each query across three child RAGs and nineteen async source workers at once, so breadth costs latency only once.
- Two-stage ranking
- Hybrid BM25 + dense retrieval fused with RRF, then a FlashRank cross-encoder rerank. It runs on CPU, so accuracy costs nothing extra.
- Answers in ~150ms when cached
- A question close enough to one already answered returns from the semantic cache with no model call at all.
- Your documents stay yours
- A shared public knowledge base plus private per-tenant documents, isolated by API key.
A small studio that ships whole products.
We build full products, not demos — every one is live or in active build, run on infrastructure we operate ourselves. Owning the whole stack is what keeps our costs down, so we can ship capable AI at prices the incumbents can't match, and we design for trust: answers you can verify, data you control.
Own products, end to end
We design, build, and run each product ourselves — on infrastructure we operate, not a reseller's.
Affordable by design
We own the whole stack, so there's no reseller margin — capable AI at prices the incumbents can't match.
Answers you can check
Grounded, cited answers and real document understanding — built so you can verify, not just hope.
Built for the real world
We start with the needs of users the industry overlooks, then build for everyone.
Gokulakannan R
Founder
I design, build and run every 26io Techlab product end to end — Evi, a personal AI assistant that plans its answer before writing it, and Kus, a retrieval API that makes small models answer reliably with citations. That means the model pipeline, the product, and the infrastructure it runs on. I'm currently researching a lightweight language model small enough to run almost anywhere and cheap enough for students and independent developers.
We're building our own lightweight AI model.
We're researching our own lightweight language model — small and cheap enough to run almost anywhere, yet tuned to punch above its weight. It's early-stage R&D, and it's where 26io is headed: an efficient model we'll offer through an affordable API, so students and developers get frontier-grade capability without frontier prices.
Start with Evi.
Our AI assistant is live and free to try. One 26io account works across Evi and Kus.
Try Evi free →