OGMA is a frontier AI lab focused on multimodal agent systems — software that sees, reasons, and acts. We build models, runtimes, and developer tools that turn screens, images, and visual documents into structured work.
We believe the most useful AI in the next decade will be the kind that meets a human halfway: it watches what the user sees, understands the task, plans a route through it, and reports back honestly when it cannot. Not autonomous, not flashy, not magic — just genuinely helpful at the boundary between perception and action.
A small family of products around a single thesis: vision is the missing API for software. Everything a developer, an analyst, or an end-user touches has a visual surface — a screen, a document, a dashboard, a chart. Making that surface programmable, without scraping, without brittle automation, and without sending sensitive data to opaque third parties — that is the problem space we are committing to.
Today we are working on:
A self-hostable runtime that takes a screenshot or an image and produces a structured task plan, then walks it through specialized agents to a grounded markdown report. Includes a safe, read-only code-agent module for repository-aware reasoning.
- Public: https://github.com/berkmec/ogma-optron
- License: MIT
- Status: MVP shipped; entering an extended dev-tool expansion phase (CLI, semantic repo intelligence, sandboxed browser agent, deepthink reasoning).
We are working on additional model families and runtimes. Details are intentionally not public yet — we will announce them when they are ready to ship.
We want OGMA to be the home of:
- Honest AI — systems that say "I don't know" instead of inventing answers, that quote evidence verbatim instead of paraphrasing it into something convenient.
- Self-hostable AI — first-class open-source products developers can run on their own infrastructure with their own credentials, not just SaaS dashboards we control.
- Composable AI — model and provider choice as a feature, not as lock-in. We do not believe the future of this industry is one vendor and one endpoint.
- Practical AI — products people actually use to finish work, not benchmarks we win on a leaderboard and never ship.
This is a long arc. We are early.
OGMA is at an early stage. If you are a fund, an angel, or a strategic partner who is interested in:
- frontier multimodal AI with a developer-first distribution
- open-core / open-source business models in the AI tooling space
- teams that ship in public and measure what they build
…we would like to hear from you.
- Reach: invest@counfield.com
- We respond personally to every introduction and will share a detailed memo, roadmap, and traction update on request.
We deliberately do not publish our roadmap or architecture publicly. If you would like to dig into the technology under NDA, write to us.
You can start with the open-source side today:
git clone https://github.com/berkmec/ogma-optron
cd ogma-optron
# follow the README — backend, frontend, and the read-only code agent- Issues and pull requests: https://github.com/berkmec/ogma-optron/issues
- General: hello@counfield.com
If you build something on top of OGMA tools, we want to see it.
The contents of this repository (the text, the mission statements, the brand assets) are released under Creative Commons Attribution 4.0 (CC BY 4.0). You are free to quote and share, with attribution.
Our open-source products are licensed individually inside their own repositories (Ogma Optron is MIT). Closed products are not covered by this license.
OGMA · counfield.com · © 2026