Check which Ollama models can run on your hardware. Detects your GPU, VRAM, RAM, and CPU, then shows a color-coded compatibility table for 370+ models.
- Interactive TUI with 4 tabs: Hardware, Compatibility, Models, Pull
- Hardware detection for NVIDIA, AMD, Apple Silicon, and CPU-only setups
- Model compatibility with color-coded verdicts (CAN RUN / DEGRADED / CAN'T RUN)
- Model browser with search and local/remote toggle
- Download models directly with progress bar
- Cross-platform single binary (Linux, macOS, Windows) -- no CGO
Download the latest binary from Releases and add it to your PATH.
go install github.com/putzeys/ollafit@latestOr clone and build:
git clone https://github.com/putzeys/ollafit.git
cd ollafit
go install .ollafitOpens a full-screen interface with 4 tabs:
| Tab | Description |
|---|---|
| 1 Hardware | Detected CPU, GPU, RAM |
| 2 Compatibility | Color-coded table for 370+ models |
| 3 Models | Browse remote or local models |
| 4 Pull | Check compatibility and download models |
| Key | Action |
|---|---|
Tab / 1-4 |
Switch tab |
j/k or Up/Down |
Navigate list |
/ |
Search (Compatibility & Models) |
q |
Cycle quantization: Q4_K_M, Q8_0, FP16 (Compatibility) |
l |
Toggle local/remote (Models) |
Enter |
Confirm / send to Pull |
Esc |
Cancel |
Ctrl+C |
Quit |
# Detect hardware
ollafit scan
# Check model compatibility
ollafit check
ollafit check --search llama
ollafit check --quant FP16
ollafit check --json
# Browse models
ollafit models
ollafit models --search llama
ollafit models --local
# Download a model (requires Ollama running)
ollafit pull llama3.2:1b| Status | Meaning |
|---|---|
| CAN RUN (green) | Model fits in GPU VRAM |
| DEGRADED (yellow) | Needs CPU/RAM offload (slower) |
| CAN'T RUN (red) | Insufficient total memory |
Create ~/.config/ollafit/config.yaml:
ollama_host: "http://localhost:11434"
model_source: "ollamadb"
vram_overhead_percent: 20.0
gpu_memory_fraction: 0.75 # Apple Silicon: % of unified memory for GPUEnvironment variables are also supported with the OLLAFIT_ prefix (e.g., OLLAFIT_OLLAMA_HOST).
- Ollama running locally for
models --localandpullcommands - GPU detection uses:
nvidia-smi(NVIDIA),rocm-smi(AMD), or Apple Silicon system APIs - Go 1.24+ only if building from source
MIT