Guide to configuring GPT4All for optimal performance and use cases.
| Platform | Location |
|---|---|
| Windows | %APPDATA%/GPT4All |
| macOS | ~/Library/Application Support/GPT4All |
| Linux | ~/.config/GPT4All |
| Platform | Location |
|---|---|
| Windows | C:/Users/<Username>/AppData/Local/nomic.ai/GPT4All/models |
| macOS | ~/Library/Application Support/nomic.ai/GPT4All/models |
| Linux | ~/.local/share/nomic.ai/GPT4All/models |
Settings → Model:
| Setting | Description | Recommended |
|---|---|---|
| Default Model | Model to load on startup | Your preferred model |
| GPU Layers | Layers to offload to GPU | Max for best performance |
| Context Length | Model context window | 4096-8192 |
| Threads | CPU threads for inference | Match CPU cores |
| Parameter | Description | Typical Value |
|---|---|---|
| Temperature | Response randomness | 0.7 |
| Max Tokens | Maximum response length | 2048 |
| Top P | Nucleus sampling | 0.9 |
| Top K | Top-K sampling | 40 |
| Repeat Penalty | Prevent repetition | 1.1 |
LocalDocs → Collections:
| Setting | Description |
|---|---|
| Name | Collection display name |
| Path | Folder path containing documents |
| Enabled | Toggle collection on/off |
| Re-index | Force re-indexing |
| Setting | Description | Default |
|---|---|---|
| Chunk Size | Text chunk size for embedding | 512 |
| Overlap | Chunk overlap | 50 |
| Embedding Model | Model for embeddings | Nomic Embed |
| Type | Extensions |
|---|---|
| Text | .txt, .md |
| Documents | .pdf, .docx |
| Presentations | .pptx |
| Spreadsheets | .csv, .xlsx |
| Code | .py, .js, .ts, .java, .cpp, etc. |
Settings → API:
| Setting | Description | Default |
|---|---|---|
| Enable API | Toggle Docker API | Off |
| Port | API listening port | 4891 |
| Host | Binding address | localhost |
# Chat completions
POST http://localhost:4891/v1/chat/completions
# Completions
POST http://localhost:4891/v1/completions
# Models
GET http://localhost:4891/v1/models
curl http://localhost:4891/v1/chat/completions \
-H "Content-Type: application/json" \
-d '{
"model": "llama-3-8b-instruct",
"messages": [
{"role": "user", "content": "Hello!"}
],
"max_tokens": 1024
}'
Settings → Hardware:
Settings → Hardware:
Settings → Hardware:
Settings → Appearance:
| Setting | Options |
|---|---|
| Theme | Light, Dark, System |
| Font Size | Small, Medium, Large |
| Sidebar | Show/Hide |
Settings → Chat:
| Setting | Description | Default |
|---|---|---|
| Show Thumbnails | Show model thumbnails | On |
| Save Chat History | Persist conversations | On |
| Export Format | Default export format | Markdown |
Create ~/.gpt4all/config.json:
{
"model_path": "/path/to/models",
"allow_download": true,
"verbose": false
}
# Custom model path
export GPT4ALL_MODEL_PATH=/path/to/models
# Disable telemetry
export GPT4ALL_TELEMETRY=0
Settings → Performance:
| Setting | Description | Recommended |
|---|---|---|
| Memory Limit | Max RAM for model | 80% of available |
| GPU Memory | Max VRAM for GPU | 80% of available |
| Thread Count | CPU threads | Match CPU cores |
| Setting | Description | Default |
|---|---|---|
| Model Cache | Cache loaded models | Enabled |
| Response Cache | Cache frequent responses | Disabled |
Settings → Advanced → Export:
Windows:
xcopy /E /I "C:\Users\%USERNAME%\AppData\Local\nomic.ai\GPT4All\models" "D:\Backup\GPT4All"
macOS/Linux:
tar -czf gpt4all-backup.tar.gz ~/Library/Application\ Support/nomic.ai/GPT4All/models
# or
tar -czf gpt4all-backup.tar.gz ~/.local/share/nomic.ai/GPT4All/models
| Issue | Solution |
|---|---|
| Model not loading | Check RAM, try smaller quantization |
| Slow performance | Enable GPU, reduce context length |
| LocalDocs not working | Re-index collection, check file formats |
| API not responding | Check port, verify Docker running |
Delete configuration file:
%APPDATA%/GPT4All/settings.json~/Library/Application Support/GPT4All/settings.json~/.config/GPT4All/settings.jsonAny questions?
Feel free to contact us. Find all contact information on our contact page.