Custom Inference Endpoint (BYOE)
Connect custom inference endpoints for feature extraction with real-time or async processing
Why do anything?
Built-in extractors may not fit your domain. Custom models need custom inference endpoints.
Why now?
ML teams have specialized models. Integration should be seamless.
Why this feature?
BYOE allows connecting any HTTP endpoint as a feature extractor. Supports real-time and async patterns with automatic retries.
How It Works
BYOE integrates custom inference endpoints into the Mixpeek feature extraction pipeline.
1
Request Formation
Format input according to endpoint schema
2
Authentication
Add auth headers/tokens from secrets
3
Invocation
Call endpoint with retry logic
4
Response Parsing
Parse output into Mixpeek document format
Why This Approach
HTTP interface enables any model deployment platform. Secrets integration keeps credentials secure.
Where This Is Used
Integration
plugin = client.plugins.byoe.create(endpoint_url="https://your-model.com/embed")