> ## Documentation Index
> Fetch the complete documentation index at: https://oxenai-eric-document-openclaw.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.
# 🗂️ Deploy From Directory
> Drop an `oxen.toml` into any directory in your repo to get a one-click Deploy Model button in the UI.
Any directory that contains an `oxen.toml` with a valid `[model]` section becomes deployable from the Oxen.ai UI. Browse to the folder and a **Deploy Model** button appears alongside the file listing. Clicking it provisions a dedicated inference endpoint using the settings in `oxen.toml`.
Below is an example of the deploy model button:
## The `oxen.toml` file
Place the file at the root of the directory you want to deploy. Only one `[model]` section is required.
```toml oxen.toml theme={null}
[model]
source = "Qwen/Qwen3-8B"
type = "lora"
```
| Field | Required | Description |
| -------- | -------- | ---------------------------------------------------------------------------------------------------------------------- |
| `source` | yes | The base model this directory was trained from. Accepts a HuggingFace ID (e.g. `Qwen/Qwen3-8B`) or an Oxen model name. |
| `type` | yes | Either `"lora"` (LoRA adapter weights only) or `"full"` (full model weights). |
## Examples
### LoRA adapter
A directory containing LoRA weights trained on top of a supported base model:
```toml oxen.toml theme={null}
[model]
source = "Qwen/Qwen3-8B"
type = "lora"
```
### Full fine-tune
A directory containing full model weights:
```toml oxen.toml theme={null}
[model]
source = "meta-llama/Llama-3.2-3B-Instruct"
type = "full"
```
### Image model
The `source` field works for any supported base model, including image and video models:
```toml oxen.toml theme={null}
[model]
source = "black-forest-labs/FLUX.1-dev"
type = "lora"
```
## The UI workflow
1. Push a directory containing model weights and an `oxen.toml` to your repository.
2. Navigate to that directory in the Oxen.ai UI.
3. Click **Deploy Model**. Oxen.ai reads `oxen.toml`, resolves the base model, and spins up a dedicated endpoint.
4. Once the deployment is live, use the returned model name with the [Inference API](/inference-api/overview) or the in-browser playground.
Below is an example of the deployment status in the workbench view after clicking Deploy Model:
Redeploying from the same directory and revision reuses the existing deployment rather than creating a duplicate.
## Common errors
| Message | Cause |
| ------------------------------------------------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `oxen.toml not found at /oxen.toml` | No `oxen.toml` in the directory you're trying to deploy. |
| `oxen.toml must contain a [model] section` | File parsed but the top-level `[model]` table is missing. |
| `oxen.toml missing required field: model.source` | `source` is not set. |
| `oxen.toml model.type must be "lora" or "full"` | `type` is set to something other than `"lora"` or `"full"`. |
| `Source model not found: ` | The `source` doesn't match any HuggingFace ID or Oxen model name Oxen.ai currently supports. Contact [hello@oxen.ai](mailto:hello@oxen.ai) to request support for a new base model. |
## The `oxen.toml` file
Place the file at the root of the directory you want to deploy. Only one `[model]` section is required.
```toml oxen.toml theme={null}
[model]
source = "Qwen/Qwen3-8B"
type = "lora"
```
| Field | Required | Description |
| -------- | -------- | ---------------------------------------------------------------------------------------------------------------------- |
| `source` | yes | The base model this directory was trained from. Accepts a HuggingFace ID (e.g. `Qwen/Qwen3-8B`) or an Oxen model name. |
| `type` | yes | Either `"lora"` (LoRA adapter weights only) or `"full"` (full model weights). |
## Examples
### LoRA adapter
A directory containing LoRA weights trained on top of a supported base model:
```toml oxen.toml theme={null}
[model]
source = "Qwen/Qwen3-8B"
type = "lora"
```
### Full fine-tune
A directory containing full model weights:
```toml oxen.toml theme={null}
[model]
source = "meta-llama/Llama-3.2-3B-Instruct"
type = "full"
```
### Image model
The `source` field works for any supported base model, including image and video models:
```toml oxen.toml theme={null}
[model]
source = "black-forest-labs/FLUX.1-dev"
type = "lora"
```
## The UI workflow
1. Push a directory containing model weights and an `oxen.toml` to your repository.
2. Navigate to that directory in the Oxen.ai UI.
3. Click **Deploy Model**. Oxen.ai reads `oxen.toml`, resolves the base model, and spins up a dedicated endpoint.
4. Once the deployment is live, use the returned model name with the [Inference API](/inference-api/overview) or the in-browser playground.
Below is an example of the deployment status in the workbench view after clicking Deploy Model:
Redeploying from the same directory and revision reuses the existing deployment rather than creating a duplicate.
## Common errors
| Message | Cause |
| ------------------------------------------------ | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `oxen.toml not found at