Local LLM on 8GB Macs: what actually fits and what it can do
On an 8GB Mac a local LLM means a small model with modest context, good for driving tools and short tasks, not deep reasoning. Here's an honest picture.
On an 8GB Mac, a local LLM means a small quantized model running with a modest context window. It won’t match a large cloud model at deep reasoning, but it’s genuinely useful for driving tools, short edits, and structured tasks — the work where a capable-enough model calling reliable tools beats a brilliant model you have to copy-paste for. ReDock is a free native macOS app that ships a local-LLM catalog tiered to your Mac’s RAM, so an 8GB machine gets a model sized to fit, with no API keys and nothing leaving the device.
What fits in 8GB
Memory is the hard constraint. Your Mac needs RAM for macOS, your browser, your editor, and the running sites — the model gets what’s left. On an 8GB machine that budget is tight, so the realistic choice is a small model in a compact quantization, run with a modest context window rather than a huge one.
That rules out the largest local models, which want far more memory than you have to spare. It does not rule out useful work. A small model that fits and stays responsive is worth more in practice than a large one that swaps to disk and stalls.
ReDock’s catalog handles the sizing for you. It tiers models to the machine’s RAM and picks one that fits, so you’re not guessing at parameter counts and quantization levels. See the built-in agent docs for how the catalog and tiers work.
What a small local model is good at
Be honest about the ceiling and the floor. A small local model is not going to write a subtle architecture proposal or reason through a tangled bug the way a large cloud model can. What it can do well:
- Drive tools. Deciding which tool to call with which arguments is a narrower task than open-ended reasoning, and small models handle it acceptably — especially against a well-designed tool surface.
- Short, structured edits. Renaming, reformatting, filling a template, drafting a paragraph.
- Routine WordPress and site work. Inserting a pattern, creating a page, activating a plugin — through ReDock’s MCP tools, the model chooses the call and the platform does the heavy lifting and verifies the result.
The pattern that works is a modest model plus reliable tools. The tools carry the correctness; the model just needs to pick the right one.
What to expect it not to do
A small local model will sometimes lose the thread on long, multi-step reasoning, especially as the context fills up. It’ll be weaker at nuance, at code it hasn’t effectively seen, and at holding a lot of state in its head. Give it a big ambiguous task and it’ll disappoint.
So scope accordingly. Hand it bounded tasks with clear success criteria and let it verify through tools. When you need deep reasoning, that’s when a cloud model earns its keep — see Claude vs local models for agent work: an honest comparison.
What driving it looks like
You pick a model from ReDock’s catalog sized for your 8GB Mac. It downloads once and runs on-device. You tell the built-in agent: “Add an about page to the demo site.” The model reads the site’s patterns, chooses the right tool calls, and ReDock inserts the page and reads it back to confirm. No API key, no network round-trip, no usage meter. For the tradeoffs of keeping everything local, see Your code never leaves your Mac: private AI for client work and the pillar, No API key AI: run a local model that can actually do things. If MCP is new to you, What is MCP? explains the tool layer.
FAQ
Will a local model on 8GB feel slow?
It depends on the model and what else is running. A model sized to fit stays responsive; one that’s too big for your free RAM will swap and stall. ReDock’s RAM-tiered catalog picks one that fits so you avoid the stall.
Can it really run WordPress tasks?
For bounded tasks, yes. Through ReDock’s MCP tools the model picks the call and the platform executes and verifies it, so correctness doesn’t rest on the model’s raw ability. Big ambiguous jobs are where you’d reach for a cloud model.
Do I need any API keys?
No. Local models run on-device with no keys and no accounts. Nothing leaves your Mac when you use one.
If you want a local model sized to your Mac that can actually drive tools, download ReDock — one download and a five-minute install.