mirror of
https://github.com/blakeblackshear/frigate.git
synced 2026-02-19 01:17:06 +03:00
Update genai.md to clarify OpenAI API spec
The docs don't make it very clear or intuitive that by supporting OpenAI API, you can essentially use any model. If someone has an API key for Claude/Anthropic, the OpenAI APi can be used (expensive use case, but just an example).
This commit is contained in:
parent
a42ad7ead9
commit
41be9fc184
@ -94,34 +94,64 @@ genai:
|
||||
model: gemini-1.5-flash
|
||||
```
|
||||
|
||||
## OpenAI
|
||||
# OpenAI and OpenAI-Compatible APIs
|
||||
|
||||
OpenAI does not have a free tier for their API. With the release of gpt-4o, pricing has been reduced and each generation should cost fractions of a cent if you choose to go this route.
|
||||
The OpenAI API specification has become a de facto standard that many AI providers and tools have adopted. This means you can use Frigate's OpenAI provider configuration with:
|
||||
|
||||
### Supported Models
|
||||
- OpenAI's own API service
|
||||
- Local model servers like Ollama or LM Studio when configured to serve via OpenAI-compatible endpoints
|
||||
- Other commercial providers that implement the OpenAI API spec (e.g., DeepSeek, Together.ai, Anthropic)
|
||||
|
||||
You must use a vision capable model with Frigate. Current model variants can be found [in their documentation](https://platform.openai.com/docs/models). At the time of writing, this includes `gpt-4o` and `gpt-4-turbo`.
|
||||
This gives you significant flexibility in choosing where to run your vision models - whether in the cloud or locally on your own hardware.
|
||||
|
||||
### Get API Key
|
||||
## OpenAI's API Service
|
||||
|
||||
To start using OpenAI, you must first [create an API key](https://platform.openai.com/api-keys) and [configure billing](https://platform.openai.com/settings/organization/billing/overview).
|
||||
If using OpenAI's official API service:
|
||||
|
||||
### Configuration
|
||||
1. [Create an API key](https://platform.openai.com/api-keys)
|
||||
2. [Configure billing](https://platform.openai.com/settings/organization/billing/overview)
|
||||
3. Use a vision-capable model like `gpt-4-vision-preview`
|
||||
|
||||
```yaml
|
||||
genai:
|
||||
enabled: True
|
||||
provider: openai
|
||||
api_key: "{FRIGATE_OPENAI_API_KEY}"
|
||||
model: gpt-4o
|
||||
model: gpt-4-vision-preview
|
||||
```
|
||||
|
||||
:::note
|
||||
## Using Alternative Providers
|
||||
|
||||
To use a different OpenAI-compatible API endpoint, set the `OPENAI_BASE_URL` environment variable to your provider's API URL.
|
||||
To use an alternative provider or local model server that implements the OpenAI API spec:
|
||||
|
||||
1. Set up your chosen provider/server (e.g., Ollama, LM Studio, DeepSeek)
|
||||
2. Configure it to serve via an OpenAI-compatible endpoint
|
||||
3. Set the `OPENAI_BASE_URL` environment variable to your provider's API URL
|
||||
4. Use the appropriate model name for your chosen provider
|
||||
|
||||
For example, with a local Ollama server serving the LLaVA model:
|
||||
|
||||
```yaml
|
||||
genai:
|
||||
enabled: True
|
||||
provider: openai
|
||||
api_key: "not-needed-for-local" # Some providers may not require an API key
|
||||
model: llava
|
||||
# Set OPENAI_BASE_URL=http://localhost:11434/v1 in your environment
|
||||
```
|
||||
|
||||
:::tip
|
||||
When using local model servers, ensure you've configured them properly to handle vision inputs through the OpenAI API compatibility layer. Check your provider's documentation for specific setup instructions.
|
||||
:::
|
||||
|
||||
## Choosing a Provider
|
||||
|
||||
Consider these factors when selecting a provider:
|
||||
- **Local vs Cloud**: Local providers like Ollama offer privacy and no usage costs, but require suitable hardware
|
||||
- **Cost**: Cloud providers typically charge per API call
|
||||
- **Performance**: Model size and hardware affect response speed
|
||||
- **Ease of Setup**: Cloud APIs are typically simpler to configure than local setups
|
||||
|
||||
## Azure OpenAI
|
||||
|
||||
Microsoft offers several vision models through Azure OpenAI. A subscription is required.
|
||||
|
||||
Loading…
Reference in New Issue
Block a user