Cognis
A self-hosted workspace where AI agents can talk, use tools, delegate work and run tasks.
View Cognis on GitHubFilip Pytloun · Software engineer
Hi, I’m Filip.
I work on edge-to-cloud systems at Lumilens, and build open-source software for AI agents.
I’ve spent more than 14 years on Linux, cloud platforms and observability, including work at two startups later acquired by Mirantis and F5.
Open-source work
Building an agent OS
Three projects I work on in my own time. Cognis brings agents and tools together; Intaris handles decisions about what they may do; Mnemory gives them continuity between conversations.
Intaris
Guardrails for agent tool calls, with policy, audit and a way to ask a person when needed.
View Intaris on GitHub
Mnemory
Long-term memory for agents and assistants, kept on your own infrastructure.
View Mnemory on GitHubAt work
Engineer at Lumilens
I’m a Principal Engineer at Lumilens, working on systems that connect devices at the edge to cloud services. Earlier, I worked on platforms and observability at F5 / Volterra, Mirantis, tcp cloud and GoodData.
Experience and backgroundLatest writing
Ideas in practice
New notes on the systems and decisions behind my work.
Sandboxes are necessary. They are not agent safety.
Isolation limits what an agent can reach. It does not tell us whether an allowed action is what the user wanted.
Writing
From the archive
Seven older articles on Linux and infrastructure, kept at their original URLs and dates. Some technical details will be outdated.