Skip to content

Knowledge that works where you are.

Kognis is a software company. We build AI and data systems that run on your phones and your servers, trained for your work and under your control, even where there is no connection.

Seven ways we put knowledge to work.

Each card is a line of work, with its name punched in. Open one to see where it runs and how we build it.

Open data

We connect public government data to the tools and AI assistants people already use, so every question gets an answer with its source.

How we build it

Runs on: Public portals, spreadsheets and archives.

  1. Map the portal and its quirks.
  2. Build a connector (an MCP server) and a clean catalog.
  3. Test it against real questions.

What you get: Data anyone can ask and verify.

Graph databases

We turn scattered records into knowledge graphs where every fact remembers where it came from, and AI can reason along the links.

How we build it

Runs on: Documents, databases and field reports.

  1. Model the people, places and events.
  2. Load every record with its provenance.
  3. Add retrieval over the graph (GraphRAG).

What you get: Answers that show their path.

Software development

Data pipelines, conversational agents on WhatsApp and Telegram, automated reports and web platforms, shaped to the problem and not the other way around.

How we build it

Runs on: Your systems, as they are today.

  1. Understand the work before writing code.
  2. Ship small, measure, adjust.
  3. Document everything and hand it over.

What you get: Software your own team can run.

Immersive experiences

Location-aware guides that know where a person is standing and what is worth telling right there, plus stories told in programmatic video.

How we build it

Runs on: Places, stories and archives.

  1. Curate and verify every story.
  2. Reason about place: what is near, what matters now.
  3. Deliver it on chat, in an app, or offline.

What you get: People who want to come back.

Android and edge apps

Native Android apps with language models that run on the phone itself: search, maps and document reading with no signal. Born in field work, where networks fail first.

How we build it

Runs on: Mid-range Android phones, tablets and edge boxes.

  1. Choose and compress the model for the device.
  2. Search your documents on the phone itself.
  3. Measure heat, battery and speed on real hardware.

What you get: Apps that work in airplane mode.

Sovereign AI

Models and data that stay on your hardware and inside your jurisdiction. No per-question bill to a distant cloud, and nothing leaves the building unless you say so.

How we build it

Runs on: Your servers, your devices, your laws.

  1. Start from open-weight models.
  2. Deploy them on infrastructure you control.
  3. Audit what goes in and what comes out.

What you get: AI you own outright.

Post-training and alignment

We teach open models your domain: curated datasets, supervised fine-tuning (SFT), reinforcement learning (GRPO), and evaluations that check the model says what it should.

How we build it

Runs on: Your manuals, protocols and language.

  1. Curate and balance the data.
  2. Fine-tune with SFT, then GRPO.
  3. Evaluate before anything ships.

What you get: A model that speaks your field.

Something elseA project that does not fit on a card? Tell us what you want to build and where it has to run.Book a call

Knowledge belongs where the work is.

The most useful AI will not all live in someone else's cloud. It will run on the phone in a field worker's pocket, on the server in a ministry's basement, on a laptop with no signal. Whoever holds the knowledge should hold the system, so we build systems our clients keep.

  • It works where people are.

    Offline, on the device, on hardware people can actually afford.

  • You own the model and the data.

    Code, weights and documentation are handed over. No lock-in by design.

  • We say what we do, plainly.

    No inflated claims. If something is a pilot, we call it a pilot.

  • Bilingual and regional by default.

    Built from Latin America, in Spanish and English from the first line.

Investors and partners: we are building Kognis as a company, not a one-off agency. If that thesis interests you, let's talk.

Book a conversation

From first call to handover.

  1. A first call

    You tell us the problem and where it has to run. We listen before we propose.

  2. A small pilot

    One narrow, useful piece, on your own data and your own hardware.

  3. Build and measure

    Short cycles, measured on real devices and real questions.

  4. Handover

    Code, models and documentation are yours. We stay on if you want us to.

Book a demo or a call.

Tell us a little about what you are working on and we will write back to find a time.

I would like

We only use these details to reply to you.