Run AI yourself
The practical path to running models on hardware you control. Not a benchmark list: these are judged on whether someone can actually operate them without a provider relationship.
4 projects, in the order the list declares.
| # | Project | Stars | Growth | Last commit | License |
|---|---|---|---|---|---|
| 1 | ollama/ollama Why it is here — The default answer to 'how do I run a model locally'. Its packaging matters more than its inference. | 178.8k | 2/14 | 3d ago | MIT |
| 2 | hiyouga/LlamaFactory Why it is here — Fine-tuning without a managed training service. | 74.2k | 2/14 | 5d ago | Apache-2.0 |
| 3 | Mintplex-Labs/anything-llm Why it is here — The application layer — retrieval and chat over your own documents. | 64.8k | 2/14 | 5d ago | MIT |
| 4 | StarTrail-org/LEANN Why it is here — Storage-efficient retrieval, which is the constraint that usually forces people back to a hosted vector service. | 12.8k | 2/14 | 18d ago | MIT |
Why it is here — The default answer to 'how do I run a model locally'. Its packaging matters more than its inference.
- Growth
- 2/14
- Last commit
- 3d ago
- Last release
- 3d ago
- License
- MIT
Why it is here — Fine-tuning without a managed training service.
- Growth
- 2/14
- Last commit
- 5d ago
- Last release
- 80d ago
- License
- Apache-2.0
Why it is here — The application layer — retrieval and chat over your own documents.
- Growth
- 2/14
- Last commit
- 5d ago
- Last release
- 5d ago
- License
- MIT
Why it is here — Storage-efficient retrieval, which is the constraint that usually forces people back to a hosted vector service.
- Growth
- 2/14
- Last commit
- 18d ago
- Last release
- 163d ago
- License
- MIT
Membership and order are curated in radar-watchlists.json. Metrics come from the radar’s own measurements — growth is sampled daily from GitHub’s star counts, and maintenance is read when this page is built. A metric with nothing behind it reads unknown rather than zero.