API Documentation
Gateway Layer
Bella-openapi
ReleasedBeyond chat completion, Bella-openapi integrates multiple AI capabilities including text vectorization, speech recognition, speech synthesis, text-to-image, and image-to-image generation, equipped with comprehensive billing, rate limiting, and resource management features. All capabilities have been thoroughly tested in large-scale production environments, ensuring stability and reliability.
Capability Layer
Bella-knowledge
UpcomingFocused on unified storage and management of knowledge, elegantly handling multiple knowledge sources such as files and Q&A pairs, providing powerful knowledge support for intelligent applications.
Bella-assistants
UpcomingAn open-source implementation compatible with OpenAI Assistants API and Responses API, breaking through native ecosystem limitations, supporting flexible switching between various vendor models, truly achieving develop once, use everywhere.
Bella-rag
UpcomingLeveraging Bella-knowledge data sources, it provides unified retrieval question-answering generation capabilities, supports various retrieval methods and RAG implementation paradigms, making AI responses more accurate and reliable.
Bella-workflow
UpcomingSimilar to DIFY but with many differentiated capabilities, such as callback mode, Groovy script support, batch processing capabilities, third-party data source registration, while delivering superior performance.
Bella-job-queue
UpcomingA centralized queue system that enables various basic capabilities to easily support batch processing mode, significantly improving processing efficiency and resource utilization.
Bella-realtime
UpcomingKnown for ultra-low latency and high flexibility, it supports free combination of different ASR, LLM, and TTS components to create the best user experience, while supporting cutting-edge features like multi-agent collaboration.
Inference Service Layer
Bella-general-infer
UpcomingSupports multiple models including LLM, ASR, TTS, Embedding, Rerank, compatible with mainstream inference backends such as Transformers, vLLM, SGLang, Faster-Whisper, providing one-stop solution for various AI inference needs.
Model Layer
Bella-whisper
UpcomingA Whisper model carefully fine-tuned with domain data, possessing excellent Simplified Chinese recognition capabilities, providing more accurate support for voice applications.