generated: '2026-08-27' method: searched source: https://raw.githubusercontent.com/mudler/LocalAI/master/backend/backend.proto http_status: 200 file: grpc/localai-backend.proto syntax: proto3 package: backend go_package: github.com/go-skynet/LocalAI/pkg/grpc/proto java_package: io.skynet.localai.backend java_outer_classname: LocalAIBackend services: - name: Backend fq_name: backend.Backend rpc_count: 52 streaming_rpc_count: 19 message_count: 114 note: >- LocalAI ships a real, first-party proto3 contract alongside its REST surface, and it is the more interesting of the two. `backend.Backend` is the interface every LocalAI inference backend implements — llama.cpp, vllm.cpp, audio.cpp, whisper, stable-diffusion and the rest all speak it — which makes this the contract a third party writes against to add a new engine to LocalAI, rather than the one a client writes against to call it. It is saved verbatim; nothing here was authored by this pipeline. It also covers capability the REST surface does not expose: fine-tuning (StartFineTune, FineTuneProgress, StopFineTune, ListCheckpoints, ExportModel), quantization (StartQuantization, QuantizationProgress, StopQuantization), distributed forwarding (Forward), and bidirectional live audio (AudioTranscriptionLive, AudioTransformStream, AudioToAudioStream). rpc_groups: - group: lifecycle rpcs: [Health, Free, LoadModel, Status, ModelMetadata, GetMetrics] - group: text rpcs: [Predict, PredictStream, Embedding, TokenizeString, Detokenize, Rerank, TokenClassify, Score] - group: image-video-3d rpcs: [GenerateImage, UpscaleImage, GenerateVideo, Generate3D, Detect, Depth] - group: audio rpcs: - AudioTranscription - AudioTranscriptionStream - AudioTranscriptionLive - TTS - TTSStream - SoundGeneration - SoundDetection - VAD - Diarize - AudioEncode - AudioDecode - AudioTransform - AudioTransformStream - AudioToAudioStream - group: recognition rpcs: [FaceVerify, FaceAnalyze, VoiceVerify, VoiceAnalyze, VoiceEmbed] - group: stores rpcs: [StoresSet, StoresDelete, StoresGet, StoresFind] - group: training rpcs: [StartFineTune, FineTuneProgress, StopFineTune, ListCheckpoints, ExportModel] - group: quantization rpcs: [StartQuantization, QuantizationProgress, StopQuantization] - group: distributed rpcs: [Forward] buf_registry: published: false note: Not published to buf.build; the .proto is distributed in the source repository only.