GitHub

Local AI No token costs Data kept private Runs offline Built in EU

Run models on your own hardware.

Local AI

AI that runs locally on your devices or servers – keeping your data out of the cloud and under your control.

Nobodywho builds local AI infrastructure that enables LLMs to run fully offline, both on-premises and directly on edge devices. Run AI whilst maintaining full control of your data and lowering costs.

Product

Run AI on your own servers or directly on your devices, kept fast and up to date by our Assurance Layer.

Edge AI

Run AI on your devices

Run more than 10,000 open-source models directly on desktops, phones, smartwatches, industrial machines, cars, drones and robots – offline, with no internet connection required.

Learn more

On-Prem AI

Run AI on your servers

Deploy AI on your own or leased GPU servers, with multi-user access, predictable costs and no per-token fees.

Learn more

Assurance Layer

Keep your local AI performing at its best

Manage, monitor, re-train and optimise AI across edge devices and on-premises servers through a single control layer.

Learn more

The numbers that matter when the model runs locally.

  • Data leaving your infrastructure0bytesNot a byte of prompts or outputs leaves your infrastructure. Sovereignty is structural, not contractual.
  • Cost per token€0Flat pricing by infrastructure or seat: a line item finance can forecast.
  • Open models10,000+Deployable on your hardware without vendor lock-in. Any GGUF from Hugging Face.
  • Language SDKs7Rust, Kotlin, Swift, Python, Dart, TypeScript, GDScript. Same features in each.

Key properties

  • Open-source inference engine, EUPL 1.2 licensed
  • On-device SDK for apps, games and embedded systems
  • On-prem inference on hardware you control
  • Orchestration across device, edge and data centre
  • Works without an internet connection
  • No per-token costs, no vendor lock-in
  • Built and maintained in Europe

Sovereign AI

Distributed compute

Open Source

You can read every line of the engine that runs your models. For regulated industries and public institutions, that is not a nice-to-have, it is the precondition for trust.

Local AI, built in Europe. Start with the docs or write and say hello.

Book a meetingWrite to us

Enterprise

Managed local AI, for your enterprise

Run AI on your own servers or directly on your devices, with the Nobodywho inference engine at the core. Our Assurance Layer keeps models fast, accurate and up to date, and our team is with you from first setup to daily operation. We make switching to local AI seamless.

Book a meetingWrite to us

Our Product

01

Edge AI

Run AI on your devices.

With our inference engine you can run more than 10,000 open-source models directly on desktops, phones, smartwatches, industrial machines, cars, drones and robots – even offline, with no internet connection required.

Our inference runtime embeds directly into your application, using each device's hardware to accelerate AI. No server required, wherever your software runs.

All modalities

LLM/VLM inference, speech-to-text, text-to-speech, voice activity detection, embeddings and reranking in one SDK.

Type-checked tool calling

Nobodywho detects tool calls and automatically triggers schema-constrained generation derived from the function signature. This ensures every call uses the expected argument names and value types.

Self-configuring deployment

Hugging Face download and cache, memory-based model selection, automatic CPU/GPU selection and memory-aware layer offloading.

Automatic context shifting

Turn-aware context shifting removes complete old exchanges while preserving system instructions, recent turns and tool-call structure.

7 native languages

All features are available across Rust, Kotlin, Swift, Python, Dart, TypeScript and GDScript.

Powered by llama.cpp & ONNX

Architecture-specific CPU instructions and GPU kernels with support for Metal, Vulkan, CUDA and numerous other backends. Supports quantized models, which reduce model size and memory requirements.

02

On-Prem AI

Run AI on your servers.

Deploy AI on your own or leased GPU servers, with multi-user access, predictable costs and no per-token fees.

Enterprise inference on infrastructure you control, built for regulated and classified environments with strict data residency requirements.

Smart deployment & inference configuration

Nobodywho benchmarks models and serving configurations for your hardware and GPU topology, then deploys the best-fit setup for your workload and desired balance of model quality, latency and throughput. Priority-aware routing and scheduling keep user requests responsive while background agents use otherwise available capacity.

Enterprise integrations

Nobodywho connects to common enterprise software such as Slack, email and CRM through ready-to-use integrations. For internal or specialized systems, we can build a custom integration, or your team can connect it directly through our API.

Governed agent runtime

Nobodywho agents connect to enterprise integrations in both directions: employees can invoke them from tools such as Slack, while agents can use connected systems to complete work. They can also run scheduled or event-driven workflows in the background, with permissions, approval gates and full auditability.

Powered by vLLM

vLLM is a high-performance open-source engine with rapid support for new models. It exposes an OpenAI-compatible API, which allows existing systems and integrations to use the local deployment with minimal integration changes.

03

Assurance Layer

Keep your local AI performing at its best.

Manage, monitor, re-train and optimise AI across edge devices and on-premises servers through a single control layer.

Evaluation and observability

Measure model quality against your own data and monitor performance in production. Visibility into inference activity, usage patterns and anomalies across your deployment.

Model management

Model selection based on your domain use cases and hardware, evaluated against your own data. We update the models as better ones become available, your hardware changes or your requirements evolve.

Hardware optimisation

Tune inference performance for your devices and servers: the right quantisation, GPU backend and serving configuration for the hardware you actually run on.

Fine-tuning and re-training

Adapt models to your domain, terminology and tasks using your own data. We fine-tune and evaluate models against your use cases, then re-train as your data and requirements evolve to maintain quality over time.

Routing and integrations

Our hybrid setup automatically routes each task to the right model – running it on your edge device when the local model is capable, or sending it to a larger model on your servers or in the cloud when needed. Ready-made connections to tools such as Slack, email and CRM; custom integrations for internal systems, or direct access through the API.

Agentic workflows

Run AI agents that carry out tasks across your tools and systems, on demand, on a schedule or in response to events. We optimise deployments for these workflows, with permissions, approval steps and audit trails that keep your team in control.

Bespoke tooling

Extend your local AI setup with tools built around the way your organisation works. We develop custom interfaces, internal applications and deployment utilities tailored to your infrastructure, giving your team practical ways to use and manage AI.

Ongoing support

Keep deployments available and up to date, backed by 24/7 support. Training and enablement that leaves your team owning the capability, beyond the contract rather than dependent on it.

Pricing

01

Getting started

Free

We get the Nobodywho tech stack up and running on your servers or edge devices.

  • Set-up on your infrastructure, free
  • Onboarding, free
  • Support for the first two months, free
Get started
02

Assurance Layer

from 500EUR / month

  • Small500 EUR / month
  • Medium1,000 EUR / month
  • Large2,000 EUR / month
  • EnterpriseContact us
Talk to us about sizing
03

Leased hardware

On request

Lease GPU servers from us and run On-Prem AI without buying hardware. Price depends on the type of GPUs.

Request a quote

Not sure which plan fits? A short call is usually enough to tell.

Book a meetingWrite to us

About

Local AI, built in Europe

Nobodywho is a Copenhagen-based company building open-source local AI. Our inference engine runs models on devices and on organisations' own servers, keeping data inside the infrastructure they control.

For teams that want help deploying and operating it, we offer managed Edge and On-Prem AI with onboarding, model selection, monitoring and support.

LocationCopenhagen, Denmark

FAQ

Questions we get asked most often.

What is Nobodywho?

Nobodywho is an open-source inference engine for running large language models on-device or on-premises, built by a Copenhagen-based team for companies that want AI without sending data outside their own infrastructure.

Is Nobodywho open source?

Yes. The core engine is open source under the EUPL 1.2 licence and available on GitHub.

Does the EU AI Act require on-premises deployment?

No. The EU AI Act does not legally mandate on-premises deployment. Nobodywho recommends it for control and simplicity, not as a compliance requirement.

Where is Nobodywho based?

Nobodywho is based in Copenhagen, Denmark, and operates within the EU.

What's the difference between Edge and Prem?

Edge runs on-device, optimised for specific hardware, and is priced per device. Prem is a shared on-premises deployment for your organisation, priced per GPU server.

How is Nobodywho's enterprise solution priced?

Edge, Prem, and Assurance are all priced on request, based on your deployment.

What does the Assurance Layer include?

Guided onboarding, 24/7 support, model selection and optimisation, and bespoke tooling, layered on top of either Edge or Prem.

Can I try Nobodywho without talking to sales?

Yes. The open-source SDK is free to run yourself, on-edge or on-prem, with no enterprise engagement required. Start with the docs.