Smart systems,
everywhere.
We take the latest open-source AI research and build intelligent systems that work for everyone — on any device, in any language, with or without internet.
Seven billion businesses
Every person is a potential business. AI agents and intelligent systems should be as accessible as a phone — not locked behind enterprise contracts and cloud dependencies.
Agents that work for you
We build AI agents that handle real tasks — answer domain questions, process documents, translate, make decisions based on your data. Not chatbots. Autonomous systems that reason about your business and act on it.
Open research, pragmatic systems
We pick from the latest open-source AI — multilingual models, edge inference, vector search, document understanding — and assemble them into maintainable systems that run anywhere.
On old maps, the unknown was marked with dragons. We build where the map ends.
The most interesting problems in technology are the ones nobody has solved yet. New models, new architectures, new ways of putting systems together — every week the landscape shifts. We work at that edge, turning what just became possible into systems that create real value.
Our goal is to enable individuals and businesses to leverage AI that was out of reach yesterday. Pragmatic, maintainable, built to last longer than the hype cycle.
Multilingual knowledge system. One container. No internet. Ask questions in any language, get answers grounded in your documents — with sources. Translate between languages. Chat. All on the device in front of you.
A student in rural Vietnam loads a textbook and asks questions in her own language. A tourist kiosk on an island answers visitors in whatever language they speak. A community center translates government forms for people who just arrived. A classroom in a mountain village becomes a language lab. One system, every language, fully offline.
Documents go in as PDFs, Word files, or plain text — parsed with spatial layout awareness so tables and complex pages don't get lost. The retrieval pipeline uses multilingual embeddings, cross-encoder reranking for precision, and swappable LLMs: Google's Gemma 4 (Apache 2.0) or Cohere's Tiny Aya (70+ languages). Add, replace, or remove documents at runtime via API. Deploys anywhere Docker runs.
A platform that treats AI like a company you can hire. Specialists for legal analysis, project management, data extraction, customer response, code review, research, accounting — each with their own brain, tools, skills, and domain knowledge. You delegate work. They execute it. A human approves the result.
Workers accept tasks through voice, text, photo, and video. They read documents, search knowledge bases, remember context across conversations, schedule their own recurring work, and delegate to each other. A legal analyst can run four perspectives in parallel and synthesize them. A research agent can chain outputs through a multi-stage pipeline. A coordinator can spin up sub-agents on demand.
Everything runs under governance — policy checks before execution, quality thresholds before anything gets persisted, full audit trails, sensitivity classification. Nothing ships without human sign-off. That's the point: AI does the work, humans own the decisions.