Lumi / Lumi vs Ollama
Comparison
Lumi and Ollama. An honest comparison.
Both run AI models on your own computer. They're built for different jobs, and the right choice depends on what you're doing.
The short answer
Choose Ollama if you want an open-source model runner on Mac, Linux or Windows, mostly use one chat at a time, and are happy to wire up your own tools. Choose Lumi if you want a complete assistant on a Windows PC with an NVIDIA RTX card: agents that read and write your documents, remember every conversation, and keep up when many things run at once.
Side by side
| Lumi | Ollama | |
|---|---|---|
| What it is | A complete AI assistant on its own engine, Lumi Bare Metal | A runner for open models, with a command line, an app and an API |
| Platforms | Windows with NVIDIA RTX (private beta) | macOS, Linux, Windows |
| Licence | Proprietary; the core is free for local use | Open source |
| Setup | One installer; models downloaded, checked and fitted to your card | Install, then pull models; tune settings yourself |
| APIs | Works with apps built for the OpenAI, Anthropic, Ollama and LM Studio APIs | Its own API and an OpenAI-compatible endpoint |
| Agents, memory, documents | Built in: agentic mode, encrypted memory, document intelligence with citations | Not included; bring your own tools |
| Many agents at once | 8.8× faster with 200 coding agents on an RTX 3080 | Requests share the model in turn |
| A single chat on a small model | About the same speed | About the same speed, sometimes slightly faster |
The 200-agent result
We ran 200 copies of unmodified Aider v0.86.2 at once against Qwen3 8B on one RTX 3080, with thinking off on both engines. Each agent wrote and then reviewed a C++20 data structure, and every result was compiled and tested. Lumi finished all 200 in 232.9 seconds; Ollama finished all 200 in 2,049.3 seconds. Same models, same tasks, same card. The full method and the other models we tested are on the benchmarks page.
The gap comes from what happens when work overlaps. With eight agents running, Lumi also answered a new priority request in about 56–66 ms on 7–8B models, where Ollama took over two seconds.
When Ollama is the better choice
- You're on a Mac or Linux. Lumi's beta is Windows with NVIDIA RTX only.
- You need open source, or want to modify the engine itself.
- You want the widest model library one command away.
- You run one chat at a time on a small model. Here the two are about equally fast.
Moving an Ollama app to Lumi
Lumi speaks the Ollama API, so an app built for Ollama can point at Lumi instead, with no code changes. The same goes for apps built for the OpenAI, Anthropic and LM Studio APIs. That's how the 200-agent test ran: Aider talked to Lumi exactly as it would talk to Ollama.
Questions
Is Lumi faster than Ollama?
When many things run at once, yes: 8.8× faster with 200 coding agents on an RTX 3080, and much quicker to answer a new request while agents work. For a single chat on a small model, the two are about the same speed.
Is Lumi open source?
No. Lumi is proprietary, and its core is free for local use. Ollama is open source.
Does Lumi run on Mac or Linux?
Not yet. The private beta is for Windows PCs with NVIDIA RTX graphics. More graphics cards and kinds of computer are next.
Private beta
The private beta opens in waves.
Windows PCs with NVIDIA RTX graphics first. The core is free for local use.