Hayz Labs · https://hayzlabs.com/reviews/top/local-apps/
reviews · top 20 list · checked 18 Oct
Top 20 apps for running AI on your own computer
Apps that download open models and run them on your own computer, so nothing you type leaves it.
Security notes for tools on this list
- Ollama. An unauthenticated memory leak (CVE-2026-7482, 9.1) was fixed in 0.17.1. Never expose its port to the internet. Checked 18 Oct: source · more
- Open WebUI. An account takeover flaw (CVE-2026-70486) in 0.9.0 and later was fixed in 0.11.0. Checked 18 Oct: source
- ComfyUI. A cross-site scripting flaw (CVE-2026-56672) was fixed in 0.28.0, and exposed servers were pulled into a cryptomining botnet in March. Never expose it to the internet. Checked 18 Oct: source
Scored ones first, highest score first; on a tie, the more widely used. Unscored ones follow, by adoption (the measure is named on each row). Open the Table to see the number that placed each one and how it was scored.
- 1llama.cppggml-org
The engine under most local AI apps; runs quantized models on CPUs, Macs and GPUs.
Best for Developers who want the engine itself, on CPUs, Macs or graphics cardsYou give up Harder to set up than Ollama or LM Studio, which run the same models
90ChecklistFree - 2JanMenlo Research
Open-source desktop chat app that runs models fully offline; a good pick if privacy comes first.
Best for People who want a free offline chat app that can also use cloud modelsYou give up Score rests on one lab speed run; far fewer GitHub stars than Ollama
90ChecklistFree - 3OllamaOllama
Free, private way to run open coding models like Qwen3.8 27B on your own computer; skip it if you need frontier quality.
Best for Developers who want a local model API, and private coding helpYou give up Slower than mlx-lm on a Mac in one test; its bigger models are cloud-only
80ChecklistFree - 4mlx-lmApple
Command-line tool for running models fast on Apple Silicon Macs.
Best for Apple Silicon Mac owners comfortable with the command line or PythonYou give up Macs only; slower than llama.cpp at long context in one test
80ChecklistFree - 5LemonadeLemonade project (AMD backed)
Free local AI server that is the natural pick for AMD graphics cards and Ryzen AI laptops.
Best for Running text, image and speech models locally on AMD hardwareYou give up The independent test covered its backends only; small project at 5.9k stars
80ChecklistFree - 6LM Studio (with Bionic agent)LM Studio
Free desktop app for running open coding models privately, now with a local agent; skip it if you need frontier quality.
Best for People who want local, private AI in a desktop app, not the terminalYou give up App is closed source; slower than Ollama on small models in one test
80ChecklistFree - 7Open WebUIOpen WebUI Inc.
Self-hosted ChatGPT-style web page for Ollama or other local servers, good for sharing at home or work.
Best for Sharing one local model setup with family or a team in a browserYou give up Licence adds a branding clause, so not standard open source; not scored
Not scored yetFree - 8ComfyUIComfy Org
Widely used free app for running open image and video models on your own computer; expect a learning curve.
Best for Trying new open image and video models early, on your own hardwareYou give up Node graph takes time to learn; exposed servers were attacked, keep it private
Not scored yetFree - 9AnythingLLMMintplex Labs
All-in-one app for chatting with your own documents using local models.
Best for Asking questions of your own files with local models, on desktop or DockerYou give up Sends usage data by default until you turn it off; not scored yet
Not scored yetFree - 10Cherry StudioCherryHQ
Desktop app that talks to many cloud and local models in one window, with assistants and document search.
Best for People who use several cloud and local models and want one windowYou give up AGPL licence; not scored yet: no recent independent test
Not scored yetFree - 11UpscaylUpscayl
Free desktop app that enlarges and sharpens photos with AI on Windows, Mac and Linux.
Best for Enlarging photos for free instead of paying for Topaz GigapixelYou give up Needs a graphics card that supports Vulkan; not scored yet
Not scored yetFree - 12LocalAIEttore Di Giacinto (mudler)
Self-hosted server that copies the OpenAI API for text, image and voice models.
Best for Developers who want to swap the OpenAI API for one on their own serverYou give up Not scored; Ollama and LM Studio are scored free options for the same job
Not scored yetFree - 13exoexo labs
Joins several Macs into one cluster to run models too big for a single machine.
Best for Owners of several Apple Silicon Macs who want to run very large modelsYou give up Best on macOS 26.2 or later; on Linux it runs on the processor only
Not scored yetFree - 14TextGen (formerly text-generation-webui)oobabooga
Long-running local chat app with many model backends and settings, for hands-on tinkerers.
Best for Hands-on tinkerers who want many backends, extensions and settingsYou give up Last change in Aug 2026; not scored; LM Studio and Ollama are scored instead
Not scored yetFree - 15LibreChatLibreChat
MIT-licensed ChatGPT-style app for teams, with sign-in, agents and MCP; works with local servers.
Best for Teams that want a self-hosted chat app with sign-in and many model providersYou give up Latest version is marked pre-release; not scored yet
Not scored yetFree - 16Open Notebooklfnovo
Self-hosted NotebookLM: add sources, ask questions with citations and make audio overviews with any model.
Best for People who want NotebookLM-style study tools on their own serverYou give up You host it yourself; not scored: the only hands-on review is from June
Not scored yetFree - 17Handycjpais
Free offline push-to-talk dictation for Windows, Mac and Linux, a real alternative to Wispr Flow.
Best for Free private dictation where your audio never leaves your computerYou give up Fewer AI rewrite features than paid apps like Wispr Flow; not scored yet
Not scored yetFree - 18MeetilyZackriya Solutions
Records, transcribes and summarises meetings on your own computer, without a bot joining the call.
Best for People who want meeting notes kept on their own computerYou give up Speaker labels need the paid PRO tier; not scored yet
Not scored yetFree - 19InvokeAIInvoke
Image app with a canvas, layers and inpainting, aimed at artists using local models.
Best for Artists who want layers and inpainting for local image modelsYou give up Results depend on the local model you load; not scored yet
Not scored yetFree - 20KoboldCppLostRuins (Concedo)
Single-file app for running local models, popular for story writing and roleplay.
Best for Story writers and roleplayers who want local models with no installYou give up AGPL licence; not scored; LM Studio is an easier free app to start with
Not scored yetFree
| # | Name | Score | What placed it | Price | Best for | You give up |
|---|---|---|---|---|---|---|
| 1 | llama.cppggml-org · Open source | 90Our checklist | About 131,600 GitHub stars: popular, which isn't the same as goodWhat it rests onTheir figure: GitHub 131.6k stars. checklist: job 2 (three independent runs: Phoronix 2026-10-05, build b10825, Vulkan and ROCm on AMD; TerminalBytes 2026-09-20, build b10809, Mac; Tom's Hardware 2026-10-15, latest build, local agent runs), model 1, value 2, privacy 2, limits 2 (new builds daily) = 9 Report a mistake | Free (open source) | Developers who want the engine itself, on CPUs, Macs or graphics cards | Harder to set up than Ollama or LM Studio, which run the same models |
| 2 | JanMenlo Research · Open source | 90Our checklist | About 45,000 GitHub stars: popular, which isn't the same as goodWhat it rests onTheir figure: GitHub 45.0k stars. checklist: job 2 (StorageReview lab run of v0.8.6, the current release, on an RTX 5000 Ada, updated 2026-10-13: 123 tok/s on Jan V3.5 4B), model 1, value 2, privacy 2 (no telemetry), limits 2 (v0.8.6 Oct 9) = 9 Report a mistake | Free (open source) | People who want a free offline chat app that can also use cloud models | Score rests on one lab speed run; far fewer GitHub stars than Ollama |
| 3 | OllamaOllama · Open source · our reviewSecurity note: see above | 80Our checklist | About 183,700 GitHub stars: popular, which isn't the same as goodWhat it rests onTheir figure: GitHub 183.7k stars. checklist: job 1 (TerminalBytes 2026-09-20, Ollama 0.32.13 on a Mac Studio M3 Ultra: 77.7 tok/s, level with llama.cpp, behind mlx-lm at 107.5; speed only), model 1, value 2, privacy 2, limits 2 (v0.40.6 Oct 16) = 8 Report a mistake | Free local; cloud Pro $20/mo, Max $100/mo | Developers who want a local model API, and private coding help | Slower than mlx-lm on a Mac in one test; its bigger models are cloud-only |
| 4 | mlx-lmApple · Open source | 80Our checklist | About 7,400 GitHub stars: popular, which isn't the same as goodWhat it rests onTheir figure: GitHub 7.4k stars. checklist: job 1 (TerminalBytes 2026-09-20, mlx-lm 0.31.3: fastest on an M3 Ultra at 107.5 tok/s; Vetted Consumer 2026-09-30: slower than llama.cpp at long context, so mixed), model 1, value 2, privacy 2, limits 2 (0.32.2 Oct 16) = 8 Report a mistake | Free (open source) | Apple Silicon Mac owners comfortable with the command line or Python | Macs only; slower than llama.cpp at long context in one test |
| 5 | LemonadeLemonade project (AMD backed) · Open source | 80Our checklist | About 5,900 GitHub stars: popular, which isn't the same as goodWhat it rests onTheir figure: GitHub 5.9k stars. checklist: job 1 (Phoronix Oct 5 2026 tested v2026.39.1 on Radeon and Strix Halo, backends only), model 1, value 2, privacy 2 (sends nothing unless asked), limits 2 (v2026.42.1 Oct 14) = 8 Report a mistake | Free (open source) | Running text, image and speech models locally on AMD hardware | The independent test covered its backends only; small project at 5.9k stars |
| 6 | LM Studio (with Bionic agent)LM Studio · Closed · our review | 80Our checklist | No public download countWhat it rests onTheir figure: no public download count. checklist: job 1 (Mozilla.ai 2026-09-17, runtime 2.28.2: trailed llama.cpp, llamafile and Ollama on small models on a Mac and a Steam Deck; speed only, and Mozilla.ai makes llamafile, a rival in the test), model 1, value 2, privacy 2 (policy: no telemetry; update checks send version, OS, IP), limits 2 (app 0.4.26 Oct 13) = 8 Report a mistake | Free; Bionic+ $20/mo; Pro $100/mo | People who want local, private AI in a desktop app, not the terminal | App is closed source; slower than Ollama on small models in one test |
| 7 | Open WebUIOpen WebUI Inc. · Open sourceSecurity note: see above | Not scored | About 156,100 GitHub stars: popular, which isn't the same as goodNotesTheir figure: GitHub 156.1k stars. no independent test of the current version in the last 90 days; licence adds a branding clause (not OSI) Report a mistake | Free (open source) | Sharing one local model setup with family or a team in a browser | Licence adds a branding clause, so not standard open source; not scored |
| 8 | ComfyUIComfy Org · Open source · our reviewSecurity note: see above | Not scored | About 138,300 GitHub stars: popular, which isn't the same as goodNotesTheir figure: GitHub 138.3k stars. held review is a first look: no recent independent test of the app itself Report a mistake | Free (open source) | Trying new open image and video models early, on your own hardware | Node graph takes time to learn; exposed servers were attacked, keep it private |
| 9 | AnythingLLMMintplex Labs · Open source | Not scored | About 67,300 GitHub stars: popular, which isn't the same as goodNotesTheir figure: GitHub 67.3k stars. no independent test in the last 90 days; telemetry is on by default (can be turned off) Report a mistake | Free (open source) | Asking questions of your own files with local models, on desktop or Docker | Sends usage data by default until you turn it off; not scored yet |
| 10 | Cherry StudioCherryHQ · Open source | Not scored | About 50,800 GitHub stars: popular, which isn't the same as goodNotesTheir figure: GitHub 50.8k stars. no independent test in the last 90 days Report a mistake | Free (open source); Enterprise edition sold separately | People who use several cloud and local models and want one window | AGPL licence; not scored yet: no recent independent test |
| 11 | UpscaylUpscayl · Open source | Not scored | About 50,300 GitHub stars: popular, which isn't the same as goodNotesTheir figure: GitHub 50.3k stars. no independent test in the last 90 days Report a mistake | Free (open source) | Enlarging photos for free instead of paying for Topaz Gigapixel | Needs a graphics card that supports Vulkan; not scored yet |
| 12 | LocalAIEttore Di Giacinto (mudler) · Open source | Not scored | About 49,600 GitHub stars: popular, which isn't the same as goodNotesTheir figure: GitHub 49.6k stars. no independent test in the last 90 days Report a mistake | Free (open source) | Developers who want to swap the OpenAI API for one on their own server | Not scored; Ollama and LM Studio are scored free options for the same job |
| 13 | exoexo labs · Open source | Not scored | About 47,900 GitHub stars: popular, which isn't the same as goodNotesTheir figure: GitHub 47.9k stars. no independent test in the last 90 days Report a mistake | Free (open source) | Owners of several Apple Silicon Macs who want to run very large models | Best on macOS 26.2 or later; on Linux it runs on the processor only |
| 14 | TextGen (formerly text-generation-webui)oobabooga · Open source | Not scored | About 47,700 GitHub stars: popular, which isn't the same as goodNotesTheir figure: GitHub 47.7k stars. no independent test in the last 90 days; last commit Aug 17 2026 Report a mistake | Free (open source) | Hands-on tinkerers who want many backends, extensions and settings | Last change in Aug 2026; not scored; LM Studio and Ollama are scored instead |
| 15 | LibreChatLibreChat · Open source | Not scored | About 45,900 GitHub stars: popular, which isn't the same as goodNotesTheir figure: GitHub 45.9k stars. no independent test in the last 90 days Report a mistake | Free (open source) | Teams that want a self-hosted chat app with sign-in and many model providers | Latest version is marked pre-release; not scored yet |
| 16 | Open Notebooklfnovo · Open source | Not scored | About 40,300 GitHub stars: popular, which isn't the same as goodNotesTheir figure: GitHub 40.3k stars; Docker 1.42M pulls. no independent test in the last 90 days (Android Authority hands-on is dated 2026-06-20) Report a mistake | Free (open source) | People who want NotebookLM-style study tools on their own server | You host it yourself; not scored: the only hands-on review is from June |
| 17 | Handycjpais · Open source | Not scored | About 34,100 GitHub stars: popular, which isn't the same as goodNotesTheir figure: GitHub 34.1k stars. not scored: the only test (Better Stack) covers 0.8.x, which was replaced by 0.9.0 on 2026-07-01, before our 90-day window; latest is v0.9.8 (Oct 3) Report a mistake | Free (open source) | Free private dictation where your audio never leaves your computer | Fewer AI rewrite features than paid apps like Wispr Flow; not scored yet |
| 18 | MeetilyZackriya Solutions · Open source | Not scored | About 31,900 GitHub stars: popular, which isn't the same as goodNotesTheir figure: GitHub 31.9k stars. no independent test in the last 90 days Report a mistake | Free (open source); PRO tier sold separately | People who want meeting notes kept on their own computer | Speaker labels need the paid PRO tier; not scored yet |
| 19 | InvokeAIInvoke · Open source | Not scored | About 28,700 GitHub stars: popular, which isn't the same as goodNotesTheir figure: GitHub 28.7k stars. no independent test in the last 90 days Report a mistake | Free (open source) | Artists who want layers and inpainting for local image models | Results depend on the local model you load; not scored yet |
| 20 | KoboldCppLostRuins (Concedo) · Open source | Not scored | About 12,000 GitHub stars: popular, which isn't the same as goodNotesTheir figure: GitHub 12.0k stars. no independent test in the last 90 days Report a mistake | Free (open source) | Story writers and roleplayers who want local models with no install | AGPL licence; not scored; LM Studio is an easier free app to start with |
Nothing matches all of that. , or try everything.
How we ranked this
Order. Scored ones first, highest score first; on a tie, the more widely used. Unscored ones follow, by adoption (the measure is named on each row).
How we score this list. Our checklist: independent evidence that it does the job (from the last 90 days), the models it runs, value, privacy and reliability. The full rules are on How we score. Open "How it was scored" on any row to see the working.
What we did not do. We did not run any of these ourselves. Every number comes from the public source linked on its row, checked on 18 Oct 2026.
Spot a mistake? A wrong rank, price or link: tell us and we will check it against the source. Corrections are dated in News.
Reviews are drafted with help from AI models, including Anthropic's Claude. Model scores come only from outside leaderboards and independent evaluations, by a written rule.