A 2019 ThinkPad, a 2018 MacBook Pro, or a first-generation Surface Book is still a real computer. The screen works. The keyboard works. What killed it as a daily driver was a browser that grew heavier every year and a cloud model that was not built for it. Local AI flips the script: give that old machine a specific job (transcription, private chat, code hints on your side project, personal doc search) and it earns its space on the desk again. We tested seven apps that run well on 16 GB, 4 to 6 physical cores, and an integrated GPU or older discrete GPU. All of them are free.
What to look for in a local AI app for an old laptop
- Runs without a GPU. A CPU-first backend keeps you honest about performance and lets integrated graphics contribute where they can.
- Small default models. 7B and 8B language models fit in 8 to 12 GB of RAM and still do useful work.
- Low background footprint. Anything that eats 30% CPU when idle is out.
- One clear job. Old machines do best when they specialise. A single-purpose app beats a jack-of-all-trades tool.
- Works alongside cloud AI. Keep ChatGPT on your main machine and use local AI on the old laptop for private or bulk tasks.
- Zero telemetry. No sending your notes to a hosted server, no analytics.
Quick comparison
| App | Best for | Free plan | Platforms | Min RAM |
|---|---|---|---|---|
| Ollama | Local model backbone | Yes | Windows, macOS, Linux | 8 GB |
| LM Studio | GUI-first chat | Yes | Windows, macOS, Linux | 8 GB |
| Whisper.cpp | Transcription | Yes | Windows, macOS, Linux | 4 GB |
| GPT4All | One-click local chat + LocalDocs | Yes | Windows, macOS, Linux | 8 GB |
| AnythingLLM | Personal knowledge base | Yes | Windows, macOS, Linux | 8 GB |
| Continue | Local code hints in VS Code | Yes | Windows, macOS, Linux | 8 GB |
| LocalSend | Cross-device file transfer without cloud | Yes | Windows, macOS, Linux | 2 GB |
| Enchanted | Native macOS chat client for Ollama | Yes | macOS | 4 GB |
The 8 apps that turn an old laptop into a local AI station
1. Ollama, the base install
Ollama runs on almost anything with 8 GB of RAM. Install once and you have qwen2.5:3b or llama3.2:3b ready in a minute. On an integrated-graphics laptop you get 10 to 20 tokens per second, which is fine for note drafting, summaries, and simple Q&A. Ollama exposes a local API that every other tool on this list can plug into.
Where it falls short: Bigger models (14B and up) push older laptops past what they can comfortably run. Stick to 3B and 7B variants.
Pricing:
- Free: Fully free and open source
- Paid: None
Platforms: Windows, macOS, Linux
Download: Ollama.com GitHub
Bottom line: Start here. Everything else in this article assumes Ollama or a similar runtime is present.
2. LM Studio, easiest GUI for the model you just downloaded
LM Studio is the friendliest chat window for an old laptop. Its “runtime” system picks the right backend for your hardware (Vulkan on old AMD, CPU on old Intel with no GPU, Metal on older Macs) and pushes tokens as fast as your machine allows. The workspace mode reads a local folder and answers questions about it, useful for old work archives you want to make searchable.
Where it falls short: Not open source. On very small SSDs (256 GB) the model catalogue fills up fast, so pick one or two models and delete the rest.
Pricing:
- Free: Free for personal use
- Paid: Workplace license for teams
Platforms: Windows, macOS, Linux
Download: LMStudio.ai
Bottom line: The chat client to install if you never plan to touch a terminal.
3. Whisper.cpp, the app that will actually justify keeping the laptop
Whisper.cpp turns hours of recorded meetings, interviews, or voice memos into transcripts overnight. Even a 2018 quad-core CPU handles the medium model on a batch job. Kick off a folder of MP3s before bed and wake up to SRTs. This alone can move an old laptop from “sitting in a drawer” to “the machine that transcribes everything.”
Where it falls short: Real-time transcription needs more grunt than an old laptop reliably provides. Use it for batch jobs.
Pricing:
- Free: Fully free and open source
- Paid: None
Platforms: Windows, macOS, Linux
Download: GitHub
Bottom line: Best single job for an old laptop. Dedicate it and never think about transcription again.
4. GPT4All, one-click local chat with LocalDocs
GPT4All ships a small default model that runs on almost any laptop and adds a “LocalDocs” panel where you drop folders and ask questions about the contents. On an old laptop this becomes a private search engine over your personal notes, PDFs, and manuals without any indexing service touching the cloud.
Where it falls short: Indexing large folders is slow on older CPUs. Keep it to a few gigabytes.
Pricing:
- Free: Fully free and open source (business licence for commercial use)
- Paid: None for personal use
Platforms: Windows, macOS, Linux
Download: GPT4All.io
Bottom line: Best “install and start using it” option for an old laptop that will live on the kitchen counter.
5. AnythingLLM, a personal Confluence for your household
AnythingLLM turns any pile of documents into a queryable knowledge base with a chat interface on top. Point it at your household paperwork, receipts, appliance manuals, or old work notes and ask real questions. It runs against local Ollama on modest hardware and keeps everything on disk.
Where it falls short: Setup is heavier than GPT4All. Best used with a plan for what goes in the knowledge base.
Pricing:
- Free: Fully free and open source (desktop app)
- Paid: Cloud offering for teams
Platforms: Windows, macOS, Linux
Download: AnythingLLM.com
Bottom line: Pick AnythingLLM if the old laptop is going to hold long-term reference material for the family or your side hustle.
6. Continue, cheap local code hints for weekend projects
Continue with a small local model (qwen2.5-coder:3b or 1.5b) runs on modest hardware and gives you inline hints and a chat sidebar in VS Code. It is not Copilot, but for a side project or a Raspberry Pi tutorial, it is more than good enough.
Where it falls short: Older integrated GPUs struggle with anything bigger than 3B. Model choice matters here more than usual.
Pricing:
- Free: Fully free and open source
- Paid: None
Platforms: Windows, macOS, Linux
Download: Continue.dev VS Code Marketplace
Bottom line: Add Continue if the old laptop is going to be your weekend coding machine. Skip if you never open an IDE on it.
7. LocalSend, the AirDrop replacement that makes the old laptop useful again
LocalSend is not strictly AI, but it is the app that keeps an old laptop wired into your daily workflow. Send files between phone, tablet, and the old laptop over the local network without any cloud. Pair with Whisper.cpp: drop a voice memo from your phone, get the transcript back on the same laptop in an hour.
Where it falls short: Both devices need to be on the same network. No cross-network transfer.
Pricing:
- Free: Fully free and open source
- Paid: None
Platforms: Windows, macOS, Linux, Android, iOS
Download: LocalSend.org GitHub
Bottom line: The glue that turns the old laptop from a standalone box into a household AI helper.
8. Enchanted, native macOS front end for older Macs
Enchanted is a Mac-native chat client for Ollama with the polish of an Apple system app. On a 2018 MacBook Pro or a 2017 iMac, Enchanted runs light, uses Metal for the model backend, and looks at home. If your old laptop is a Mac, this is the client to install.
Where it falls short: macOS only. No Windows or Linux build.
Pricing:
- Free: Fully free and open source
- Paid: None
Platforms: macOS
Download: Mac App Store GitHub
Bottom line: Best macOS-native chat client for older Macs. Skip if the old laptop runs Windows or Linux.
How to pick
- If the old laptop will do one thing (transcription): Whisper.cpp, done.
- If it will be the household knowledge base: AnythingLLM + Ollama.
- If it will replace your desktop chat habit: LM Studio (Windows/Linux) or Enchanted (Mac), both on Ollama.
- If it will be your coding side machine: Continue with
qwen2.5-coder:3b. - If you want to make the old laptop a file relay between your phone and main laptop: LocalSend + Whisper.cpp is the combo.
- If you have never done any of this: install GPT4All first. It works out of the box.
FAQ
What is the minimum spec for local AI on an old laptop?
8 GB of RAM, a 64-bit CPU, and 20 GB of free disk. Anything above that is a bonus.
Which model runs best on old hardware?
qwen2.5:3b and llama3.2:3b are the sweet spot for general chat. qwen2.5-coder:1.5b is surprisingly useful for a coding side machine.
Does an old GPU help at all?
A 2018-era discrete GPU with 4 GB of VRAM helps with Stable Diffusion and small model inference. For most language models on an old laptop, CPU or integrated graphics is what you actually use.
Should I upgrade the laptop’s SSD first?
If the SSD is under 256 GB, yes. Models eat disk, and swapping models around is much less fun on a full drive.
Is a 2017 MacBook Pro still worth reviving with local AI?
Yes for transcription and small-model chat. It is comfortable running qwen2.5:3b at a readable speed.
