Skip to main content

llmProvider

@codebolt/plugin-sdk

Variable: llmProvider

const llmProvider: {
  onCompletionRequest: (callback: (req: LlmCompletionRequest) => void) => void;
  onLoginRequest: (callback: (req: LlmLoginRequest) => void) => void;
  onStreamRequest: (callback: (req: LlmStreamRequest) => void) => void;
  register: (manifest: LlmProviderManifest) => Promise<LlmProviderRegisterResponse>;
  sendChunk: (requestId: string, chunk: any) => void;
  sendError: (requestId: string, error: string) => void;
  sendReply: (requestId: string, response: any, success: boolean) => void;
  unregister: (providerId: string) => Promise<LlmProviderResponse>;
};

Defined in: CodeBolt/packages/pluginSdk/src/modules/llmProvider.ts:166

Type Declaration​

NameTypeDescriptionDefined in
onCompletionRequest()(callback: (req: LlmCompletionRequest) => void) => voidSubscribe to incoming non-streaming completion requests. Reply with sendReply(requestId, response) or sendError(requestId, error).CodeBolt/packages/pluginSdk/src/modules/llmProvider.ts:195
onLoginRequest()(callback: (req: LlmLoginRequest) => void) => voidSubscribe to incoming login requests (triggered by the UI login button). The plugin should run its authentication flow (e.g. OAuth) and then call sendReply(requestId, { authenticated: true }) or sendError().CodeBolt/packages/pluginSdk/src/modules/llmProvider.ts:215
onStreamRequest()(callback: (req: LlmStreamRequest) => void) => voidSubscribe to incoming streaming completion requests. Stream tokens with sendChunk(requestId, chunk), then finalize with sendReply(requestId, finalResponse) or sendError(requestId, error).CodeBolt/packages/pluginSdk/src/modules/llmProvider.ts:205
register()(manifest: LlmProviderManifest) => Promise<LlmProviderRegisterResponse>Register this plugin as a custom LLM provider on the server. After registration, the provider appears in the provider list and can be selected by users like any built-in provider.CodeBolt/packages/pluginSdk/src/modules/llmProvider.ts:172
sendChunk()(requestId: string, chunk: any) => voidSend a streaming chunk for an in-flight stream request. The chunk shape should match multillm StreamChunk (id, model, choices: [{delta: {content, ...}}]).CodeBolt/packages/pluginSdk/src/modules/llmProvider.ts:225
sendError()(requestId: string, error: string) => voidSend an error for a completion or stream request.CodeBolt/packages/pluginSdk/src/modules/llmProvider.ts:252
sendReply()(requestId: string, response: any, success: boolean) => voidSend the final reply for a completion or stream request. For non-streaming: this is the only message you send. For streaming: send after all chunks have been emitted.CodeBolt/packages/pluginSdk/src/modules/llmProvider.ts:239
unregister()(providerId: string) => Promise<LlmProviderResponse>Unregister this plugin's provider.CodeBolt/packages/pluginSdk/src/modules/llmProvider.ts:183