Since 2008 · Ljubljana, Slovenia

The runtime and the substrate
for the next generation of AI.

Two products. One thesis. We are building the infrastructure that lets AI systems think longer, remember more, and act at scale — with provenance, cost, and exactly-once guarantees the current stack cannot offer.

Invest with us See the products

30+

years of AI & systems research

flagship products in active development

year partnership with DELO, Slovenia's largest media house

Flagship products

Two pillars of the AI stack we’re building.

Independent, complementary. One executes; the other remembers.

Archimed · Distributed AI runtime

Give AI agents a place to stand.

A streaming shell over AI kernels, meshed across your machines. Archimed compiles ordinary async Python into a durable graph and executes it across a mesh of AI kernels driving the real Claude Code and Codex CLIs — with provenance, cost accounting, and exactly-once effects.

Durable graph execution. Long-running research workflows survive restarts, crashes, and reassignment across nodes.
A mesh of real agents. Nodes find each other through agora and talk over encrypted QUIC — no open ports, no simulated tool use.
Exactly-once effects. Idempotency and provenance are first-class primitives, not afterthoughts.
Cost as a signal. Every token, every kernel-second is measured, attributable, and budgetable.

View on GitHub → “Give me a place to stand, and I will move the earth.” — Archimedes

$ arc up research.py
✓ compiled → durable graph · 14 nodes
agora ljubljana-1 · ljubljana-2 · fra-1
▸ kernel(claude · "reviewer") running
▸ kernel(codex · "builder") running
✓ effects exactly-once · journal 1.2k events
$ arc top
  /cost 0.42 · /energy 62 Wh
  mother: watching · all services green
█

MindMem · Cognitive substrate

A living memory for machine minds.

Most “memory” in software is an archive — a wiki waits to be read, an index waits to be queried. A MindMem tissue is a small living memory body: connect it to compute and it begins to breathe — its cells activate, react to each other, dream, and forget.

Cells, not rows. Knowledge lives as cells in a tissue that consolidates, decays, and creatively recombines — structure the transformer never had.
It feeds itself. Point it at YouTube channels, X accounts, RSS feeds, transcripts — the tissue ingests and grows on its own.
Yours, entirely. Private by design — never shared, never trained on, exportable. Own it locally over MCP, REST, or CLI.
Farm or Lab. Grown at scale in the Farm, or hand-curated in the Lab with premium dream cycles and dedicated curators.

View on GitHub → Try the live demo →

The thesis

Foundation models plateau. Systems around them do not.

The next decade of AI value comes from what surrounds the model, not the model itself.

durability

Execution needs durability.

Agent workflows measured in hours or days need the same primitives distributed databases needed 20 years ago: replay, idempotency, exactly-once. Archimed provides them.

structure

Memory needs structure.

A wall of embeddings is not a mind. MindMem gives knowledge shape — cells, tissue, chemistry — and a process for consolidation, forgetting, and creative recombination.

cost

Cost is the new latency.

The teams that win will treat compute budgets as a control signal. Both products measure and attribute cost end-to-end — from graph node to CLI invocation to token.

timing

The window is now.

Every serious lab is racing to build private versions of what we open. First-mover access to Archimed and MindMem is available — before it isn’t.

Track record

Not a first attempt. A 30-year one.

We have been shipping applied AI systems since before “AI” was a fundable word.

2008–present

DELO d.o.o. — strategic technology partner

16 years running the AI, search, and publishing infrastructure for Slovenia’s largest media company: 2M+ articles indexed with a custom Slovenian stemmer, Whisper-based video transcription, GPT-based summarization, and multi-audience content transformation.

2020

DeepWriter — first on-device GPT for iOS

A GPT-based predictive writing app running 100% on-device via CoreML. Custom datasets, automatic training pipelines, live model updates via S3. We were doing on-device LLMs before it was a category.

2015

Featured by Apple — Creative Writer

Creative Writer, our predictive-intelligence iOS app, was featured by Apple in Best New Apps. Backed by ML models trained on curated literature corpora.

Throughout

Deep expertise across the stack

Invest & cooperate

We’re raising — and we’re looking for partners.

Archimed and MindMem are moving from research into product. We’re opening conversations with investors, design partners, and enterprise pilots.

For investors

Seed / strategic. Two products at the frontier, a proven team, and a clear commercialization path through applied AI infrastructure.

Talk to us about investing

For design partners & pilots

Publishing, research organizations, and platform teams: run your durable agent workflows on Archimed, or pilot MindMem as your memory substrate. Early access, direct line to the team.

Explore a partnership