* Send a prompt through the appropriate backend for the given model. * - Local models (browser_local, llama_cpp_local): routed directly, no inference params * - Remote models: fetches per-model params and delegates to sendApi.fetchResponse
(
modelName: string,
preprocessResult: PreProcessorResult,
token?: string,
enableStreaming: boolean = false,
onStreamChunk?: (chunk: string) => void,
onReasoningChunk?: (chunk: string) => void
)
| 678 | * - Remote models: fetches per-model params and delegates to sendApi.fetchResponse |
| 679 | */ |
| 680 | public async sendPrompt( |
| 681 | modelName: string, |
| 682 | preprocessResult: PreProcessorResult, |
| 683 | token?: string, |
| 684 | enableStreaming: boolean = false, |
| 685 | onStreamChunk?: (chunk: string) => void, |
| 686 | onReasoningChunk?: (chunk: string) => void |
| 687 | ): Promise<ModelOutput> { |
| 688 | // Resolve model → server |
| 689 | let modelsResponse = this.listModels(); |
| 690 | let model = modelsResponse.models.find(m => m.name === modelName); |
| 691 | if (!model) { |
| 692 | modelsResponse = await this.fetchModels(); |
| 693 | model = modelsResponse.models.find(m => m.name === modelName); |
| 694 | } |
| 695 | if (!model) throw new Error(`Model '${modelName}' not found in available models`); |
| 696 | |
| 697 | const serverAddress = model.server; |
| 698 | |
| 699 | // Decision models: the prompt is a System One question/request, the images are the state |
| 700 | if (model.decision && !this.isLocalModel(serverAddress)) { |
| 701 | // Hosted decision model (cloud Jev): the server answers; images ride inside the state |
| 702 | if (serverAddress.includes('api.observer-ai.com') && token) { |
| 703 | this.optimisticUpdateQuota(1); // a decision costs 1 credit, an LLM call 4 |
| 704 | } |
| 705 | const images = (preprocessResult.images ?? []).map(imageDataUrl); |
| 706 | const request = buildCloudRequest(modelName, preprocessResult.modifiedPrompt, images); |
| 707 | const { fetchDecision } = await import('./sendApi'); |
| 708 | const answers = await fetchDecision(serverAddress, request, token); |
| 709 | return { response: null, decision: cloudDecision(answers, request.single) }; |
| 710 | } |
| 711 | if (model.decision) { |
| 712 | if (model.status !== 'loaded') { |
| 713 | throw new Error('Local model not loaded. Please load it from the Add Model panel.'); |
| 714 | } |
| 715 | const images = (preprocessResult.images ?? []).map(imageDataUrl); |
| 716 | const prepared = prepareDecision(preprocessResult.modifiedPrompt, images); |
| 717 | const logits = serverAddress === ModelManager.LLAMA_CPP_LOCAL |
| 718 | ? await NativeLlmManager.getInstance().decide(prepared.branches) |
| 719 | : await GemmaModelManager.getInstance().decide(prepared.branches); |
| 720 | return { response: null, decision: assembleDecision(prepared, logits) }; |
| 721 | } |
| 722 | |
| 723 | // Build messages from preprocessResult |
| 724 | const hasImages = preprocessResult.images && preprocessResult.images.length > 0; |
| 725 | let content: any = preprocessResult.modifiedPrompt; |
| 726 | if (hasImages) { |
| 727 | content = [ |
| 728 | { type: 'text', text: preprocessResult.modifiedPrompt }, |
| 729 | ...preprocessResult.images!.map(img => ({ |
| 730 | type: 'image', |
| 731 | image: imageDataUrl(img), |
| 732 | })), |
| 733 | ]; |
| 734 | } |
| 735 | const messages = [{ role: 'user', content }]; |
| 736 | |
| 737 | // Local models: route directly, they manage their own settings |
no test coverage detected