Dockery/Trilium
1
0
Fork 0
mirror of https://github.com/zadam/trilium.git synced 2025-11-08 22:35:50 +01:00
Code Issues Packages Projects Releases Activity

refactor(llm): streamline chat response handling by simplifying content accumulation and removing unnecessary thinking content processing

Browse Source
This commit is contained in:
perf3ct
2025-06-02 23:25:15 +00:00
parent aad92b57c7
commit e7e04b7ccd
4 changed files with 87 additions and 398 deletions
Download Patch File Download Diff File Expand all files Collapse all files

15
apps/server/src/services/llm/chat/rest_chat_service.ts
View File

@@ -231,21 +231,16 @@ class RestChatService {
if (data) {
message.content = data;
// Handle accumulation carefully - if this appears to be a complete response
// (done=true and data is much longer than current accumulated), replace rather than append
if (done && data.length > accumulatedContentRef.value.length && data.includes(accumulatedContentRef.value)) {
// This looks like a complete final response that includes what we've accumulated
accumulatedContentRef.value = data;
} else {
// Normal incremental accumulation
accumulatedContentRef.value += data;
}
// Simple accumulation - just append the new data
accumulatedContentRef.value += data;
}
// Only include thinking if explicitly present in rawChunk
if (rawChunk && 'thinking' in rawChunk && rawChunk.thinking) {
message.thinking = rawChunk.thinking as string;
}
// Only include tool execution if explicitly present in rawChunk
if (rawChunk && 'toolExecution' in rawChunk && rawChunk.toolExecution) {
const toolExec = rawChunk.toolExecution;
message.toolExecution = {
@@ -262,7 +257,7 @@ class RestChatService {
// Send WebSocket message
wsService.sendMessageToAllClients(message);
// Send SSE response
// Send SSE response for compatibility
const responseData: any = { content: data, done };
if (rawChunk?.toolExecution) {
responseData.toolExecution = rawChunk.toolExecution;