Previously we'd use the general openai client, even if the image
generation model has a different api key and base url set.
This change uses the openai config of the image generation models when
set. Otherwise it fallbacks to use the first openai api provider set
This change adds the ability to use OpenAI, Azure OpenAI or any embedding model exposed behind an OpenAI compatible API (like Ollama, LiteLLM, vLLM etc.).
Khoj previously only supported HuggingFace embedding models running locally on device or via HuggingFaceW inference API endpoint. This allows using commercial embedding models to index your content with Khoj.
Previously if the call to this tool AI failed, the API call would
non-gracefully fail on server. This would leave the client hanging in
a wierd state (e.g with spinner running on web app with no indication
of issue).
Also do not show filters in query debug log lines when no filters in query
- Background
Access to the clipboard API is disabled by certain browsers in non
localhost http scenarios for security reasons.
So the copy API key button doesn't work when khoj is self-hosted
with authentication enabled at a non localhost domain
- Change
This change enables copying API keys by manual text highlight + copy
if copy button is disabled
Resolves#1070
This change makes Automations (and possibly other entrypoints) use the configured OpenAI-compatible server if that has been set. Without this change it tries to use the hardcoded OpenAI provider.
- Just say using default config. This old khoj.yml settings mechamism
isn't standard, so not having a configured khoj.yml isn't a concern
- Deep link to desktop download instead of the whole download page as
android etc. are also on it, which don't help with syncing docs
OpenAI chat models deployed on Azure are (ironically) not OpenAI API compatible endpoints.
This change enables using OpenAI chat models deployed on Azure with Khoj.
The github integration has not been tested and may still be broken.
This change at least makes it easier to add repositories without
needing a PAT token if/when it does work.
- Translated comments from French to English for better accessibility and understanding.
- Updated CSS comment for loading animation to reflect the change in language.
- Enhanced code readability by ensuring consistent language usage across multiple files.
- Improved user experience by clarifying the purpose of various functions and settings in the codebase.
Fix AttributeError: 'NoneType' object has no attribute 'model_extra'
* cost = chunk.usage.model_extra.get("estimated_cost", 0) if hasattr(chunk, "usage") and chunk.usage else 0 # Estimated costs returned by DeepInfra API
- Encode article urls in filename indexed in Khoj KB
Makes it easier for humans to compare, trace retrieval performance
by looking at logs than using content hash (which was previously
explored)
