Your organisation's knowledge graph and agents, built with you on Semurg.
We design and build an organisation's enterprise in a box: its databases, object stores and documents brought into one knowledge graph with an ontology, and persistent agents that work on it, all on hardware you run. Engagements are priced on request. The Semurg licences for the nodes underneath are bought separately, at the listed per-node prices.
What we build
All your data, in one store
We connect your existing databases, object stores and document stores to Semurg and keep them in step with it, so records, files and documents from every system sit in one store you run. Each source gets a connector, configured or built for the systems you have. The originals stay where they are.
The knowledge graph and ontology
People, customers, contracts, assets and decisions become entities with typed relations, each fact cited to the record or passage it came from. The ontology builds itself from what is stored; we work with your people to name it, check it and settle the classes that matter.
A digital twin of the organisation
The graph becomes a live model of how the organisation works: who owns what, which system holds which record, what changed and when. People and agents ask the same model and get the same cited answers.
Persistent agent swarms
Agents built for your organisation's own work. Each is a supervised process on the Erlang virtual machine (the BEAM): thousands run side by side on one node, each keeps its memory in the graph, and an agent that fails restarts from that memory instead of starting over.
Models on your hardware
The embedding and retrieval models run on the nodes' own processors, so your data is not sent to a hosted model. If you choose to use an external model for a task, that is your call and we set it up the way you decide.
Applications for your teams
Applications built on Semurg for your organisation, using the same memory routes any developer uses. They stay private to you and are never part of the public release.
Agents on the BEAM, their memory on the same substrate
Not one chatbot over a vector store: thousands of long-lived agents, one per customer, case, asset or process, each an isolated process on the BEAM, the Erlang virtual machine Semurg's engine is written for. Each agent writes what it learns into the organisation's knowledge graph and reads the graph to decide what to do next, on the same Semurg nodes it runs beside. When it needs to reason it calls a model: on the node, or one you choose outside it.
Isolated
An agent shares nothing with its neighbours. One that fails on a bad document stops alone, never the agent next to it.
Fault tolerant
Every agent runs under a supervisor that restarts it, with a ceiling so a persistent failure escalates instead of looping. A restarted agent reads its memory back from the graph and carries on.
Massively concurrent
A BEAM node runs hundreds of thousands of processes across every core, so a swarm of thousands of agents is an ordinary load, not a cluster of containers.
Build it yourself today: write the swarm in Elixir with Jido or plain OTP, one memory scope per agent, starting from the complete program on the agents page; any other language reaches the same memory over MCP or HTTP. Or build it with us, including agents running inside the node's own virtual machine. The design, end to end: an enterprise in a box.
How an engagement runs
1. Scoping call
Your systems, the questions your people need answered, and the size and sensitivity of the data. We say which connectors exist today and which we would build. No charge.
2. Pilot
One area of the organisation: its sources brought in, its graph built, one agent working on it, run on your hardware and measured there.
3. Roll-out
The remaining sources and agents, and the people who will run the system trained on it as it grows.
4. Hand-over
Your team runs it. Every node takes free updates from the Marketplace like any other node, and optional support continues if you want us there.
Pricing
Engagements are priced on request, after the scoping call, because every organisation's systems differ. The nodes underneath run the same Semurg everyone gets: their licences are bought separately at the per-node prices on the pricing page (the first node is free), and support is optional.
Said plainly: Semurg runs on Intel and AMD (x86_64) machines, your own or rented: Linux directly, Windows through WSL2 or Docker; a Mac version is coming. Which connectors already exist for your systems and which would be built is named in the scoping call, never promised here. We do not name the organisations we work with.
Or write to semurg@semurg.org.