lilbee / how it compares
lilbee vs Ollama, LM Studio, and vLLM. Where each one fits, and what you actually get for a comparable-size install.
lilbee is a full alternative to a model runner: it finds, runs, and manages the models itself, with a search engine over your own files on top, and it scales that whole stack across every GPU in the machine. Already on Ollama or LM Studio? It also runs on top of them.
the short version
Of the four, lilbee is the only one built around retrieval, and the only one that scales the whole stack, chat, embedding, vision, and reranking, across every GPU in the machine behind a load-balancing router.
full comparison
Every claim below links to its source: the other projects' own docs, and lilbee's.
| lilbee | LM Studio | Ollama | vLLM | |
|---|---|---|---|---|
| Primary focus | local search, chat, and serving across your GPUs | desktop app to run and chat with models | local model runner with a growing ecosystem | high-throughput GPU serving |
| Runs local models | ✓ | ✓ | ✓ | ✓ |
| Search your own files, with citations | ✓ full RAG pipeline, inline per-line citations | per-session doc attachment (RAG, document-level citation) | — | — |
| Chat, embedding, vision, rerank as one managed fleet | ✓ all four, coordinated | chat, embed, vision (no rerank), loaded individually | chat, embed, vision (no rerank), loaded individually | each supported, but one model per server |
| Multi-GPU model placement | ✓ VRAM-aware tensor-split | ✓ GPU selection + tensor parallelism (CUDA) | ✓ auto multi-GPU offload | ✓ tensor + pipeline parallel |
| Scales the whole stack, not just one model | ✓ per-GPU replicas + load-balancing router | — | — | one model per server |
| Built for many-user throughput at scale | ✓ a data-parallel replica per GPU, requests load-balanced | — | limited | ✓ this is its job |
| Web crawler built in | ✓ built in | — | — | — |
| Long-term memory (opt-in) | ✓ opt-in | — | — | — |
| Interfaces | TUI, CLI, MCP, REST, Python, Obsidian GUI | desktop GUI, lms CLI, Python + TS SDKs, REST, MCP client | desktop GUI, CLI, REST, Python/JS libs | API server |
| Use your existing Ollama / LM Studio / cloud as a backend | ✓ how | — | — | — |
install size
Download sizes in decimal GB/MB, measured from each project's own release artifacts.
| macOS | Windows | Linux | What you get | |
|---|---|---|---|---|
| lilbee (Metal / Vulkan) | 286 MB | 303 MB | 422 MB | the whole stack: search engine, crawler, servers, TUI, model runner, fleet manager |
| lilbee (CUDA, NVIDIA) | n/a | 633 MB | 1.20 GB | the same, with the faster CUDA runtime |
| Ollama | 164 MB | 1.43 GB | 1.44 GB | a model runner, fetches its runtimes separately |
| LM Studio | 569 MB | 617 MB | 1.10 GB | a desktop app (Electron) |
| vLLM | n/a | n/a | multi-GB | a Python + CUDA serving engine |
Even lilbee's CUDA build stays under Ollama's, and it's the whole stack, not just a model runner.
where each one fits
lilbee
A local search engine with the model runner built in. It runs a whole fleet, chat, embedding, vision, and reranking, across your GPUs to answer from your own files, with citations back to the source line.
LM Studio
A polished desktop app for running and chatting with a model, with per-session document attachment and a growing developer SDK surface.
Ollama
A local model runner with a strong model library and a simple pull-and-run flow. It serves chat, embedding, and vision models, one request at a time.
vLLM
A serving engine built to push one model to many users at maximum throughput, with tensor and pipeline parallelism across a cluster.
questions
Is lilbee an alternative to Ollama or LM Studio?
Yes. lilbee is a complete model manager: it browses Hugging Face, downloads models, gives each a role, and runs them on Metal, Vulkan, or CUDA, so it can replace Ollama or LM Studio outright. It also does more, with a search engine over your own files, a web crawler, and an MCP server, and if you would rather keep your current setup it runs on top of Ollama or LM Studio instead.
How is lilbee different from vLLM?
vLLM serves one model to many users at high throughput. lilbee runs a whole local fleet, chat, embedding, vision, and reranking, on your own machine to answer questions from your files, with retrieval and citations built in.
Is lilbee bigger to install than Ollama?
No. lilbee's CUDA build is about the same size as Ollama's, roughly 1.1 GB on Linux and 0.6 GB on Windows, next to Ollama's 1.4 GB GPU download. The difference is that lilbee bundles a search engine, a web crawler, servers, and a terminal UI, not just a model runner.
Can I use my existing Ollama or LM Studio models?
Yes. Point lilbee at a running Ollama or LM Studio and its models appear in the same catalog and role pickers alongside lilbee's own, read-only, so their lifecycle stays in the app you already use.
Does lilbee support multiple GPUs?
Yes. It sizes each model with gguf-parser and tensor-splits a chat model across the fewest GPUs that fit, with the embedder, reranker, and vision models placed alongside it behind a load-balancing router.