Get started
Tests, boilerplate, explanations, and single-file changes.
MODELS TO TRY
20–24B
MEMORY CHECK
32GB needed
16GB more needed
HARDWARE OPTIONS
LOCAL AI ADVISORWhat do you want local AI to do?
Start with the job. We’ll turn it into a model size, a memory target, and hardware that can run it.
1
Choose the job
2
Describe your machine
3
Compare your options
Upgrade recommended
For local coding, start with 32GB.
1Which model should I use?
20–24B to 30–35B
2Do I have enough memory?
More memory needed
3Will it respond quickly?
Target: 20+ tokens/sec
Responsive enough for live iteration.
Compare your three paths
Pick a path. If your machine covers it, test the speed before buying anything.
Recommended
BEST BALANCEA credible daily copilot for scoped implementation.
MODELS TO TRY
30–35B
MEMORY CHECK
48GB needed
32GB more needed
HARDWARE OPTIONS
Extra headroom
For harder work, longer context, and more reliable tool use.
MODELS TO TRY
70–72B
MEMORY CHECK
96GB needed
80GB more needed
HARDWARE OPTIONS
| EXPERIENCE | MODEL TO TRY | MEMORY CHECK | HARDWARE OPTIONS |
|---|---|---|---|
Get started Tests, boilerplate, explanations, and single-file changes. | 20–24B | 32GB needed 16GB more needed | |
RecommendedBEST BALANCE A credible daily copilot for scoped implementation. | 30–35B | 48GB needed 32GB more needed | |
Extra headroom For harder work, longer context, and more reliable tool use. | 70–72B | 96GB needed 80GB more needed |
Model links go to Unsloth or the original Hugging Face page. Hardware links go to official stores; prices vary by configuration.