diff options
| -rw-r--r-- | backend/app/main.py | 129 | ||||
| -rw-r--r-- | backend/app/schemas.py | 29 | ||||
| -rw-r--r-- | backend/app/services/debate.py | 371 | ||||
| -rw-r--r-- | frontend/src/components/Sidebar.tsx | 695 | ||||
| -rw-r--r-- | frontend/src/store/flowStore.ts | 27 |
5 files changed, 1236 insertions, 15 deletions
diff --git a/backend/app/main.py b/backend/app/main.py index 304c74f..89c5dd0 100644 --- a/backend/app/main.py +++ b/backend/app/main.py @@ -8,9 +8,10 @@ from fastapi.middleware.cors import CORSMiddleware from fastapi.responses import StreamingResponse, FileResponse from fastapi import UploadFile, File, Form from pydantic import BaseModel -from app.schemas import NodeRunRequest, NodeRunResponse, MergeStrategy, Role, Message, Context, LLMConfig, ModelProvider, ReasoningEffort, CouncilRunRequest +from app.schemas import NodeRunRequest, NodeRunResponse, MergeStrategy, Role, Message, Context, LLMConfig, ModelProvider, ReasoningEffort, CouncilRunRequest, DebateRunRequest, DebateJudgeMode from app.services.llm import llm_streamer, generate_title, get_openai_client, get_anthropic_client, resolve_provider from app.services.council import council_event_stream +from app.services.debate import debate_event_stream from app.auth import auth_router, get_current_user, get_current_user_optional, init_db, User, get_db from app.auth.utils import get_password_hash from dotenv import load_dotenv @@ -584,6 +585,132 @@ async def run_council_stream( ) +@app.post("/api/run_debate_stream") +async def run_debate_stream( + request: DebateRunRequest, + user: str = DEFAULT_USER, + current_user: User | None = Depends(get_current_user_optional), +): + """ + Run a multi-round LLM Debate and stream SSE events. + """ + resolved = resolve_user(current_user, user) + username = resolved.username if resolved else DEFAULT_USER + + # Merge incoming contexts + raw_messages = [] + for ctx in request.incoming_contexts: + raw_messages.extend(ctx.messages) + if request.merge_strategy == MergeStrategy.SMART: + final_messages = smart_merge_messages(raw_messages) + else: + final_messages = raw_messages + execution_context = Context(messages=final_messages) + + # Extract images from attached files + images, non_image_file_ids = extract_image_attachments(username, request.attached_file_ids) + + openrouter_key = get_user_api_key(resolved, "openrouter") + + # Build LLMConfig + attachments + tools for each debate member + member_configs: list[LLMConfig] = [] + attachments_per_model: list[list[dict] | None] = [] + tools_per_model: list[list[dict] | None] = [] + + for member in request.debate_models: + provider = resolve_provider(member.model_name) + provider_str = provider.value + api_key = get_user_api_key(resolved, provider_str) + + config = LLMConfig( + provider=provider, + model_name=member.model_name, + temperature=member.temperature if member.temperature is not None else request.temperature, + system_prompt=request.system_prompt, + api_key=api_key, + reasoning_effort=member.reasoning_effort if member.reasoning_effort is not None else request.reasoning_effort, + enable_google_search=member.enable_google_search if member.enable_google_search is not None else request.enable_google_search, + ) + member_configs.append(config) + + # Prepare provider-specific file attachments + tools: list[dict] = [] + attachments: list[dict] = [] + scoped_file_ids = resolve_scoped_file_ids(username, request.scopes, non_image_file_ids) + + if provider == ModelProvider.OPENAI: + vs_ids, debug_refs, filters = await prepare_openai_vector_search( + user=username, + attached_ids=non_image_file_ids, + scopes=request.scopes, + llm_config=config, + ) + if not vs_ids: + try: + client = get_openai_client(config.api_key) + vs_id = await ensure_user_vector_store(username, client) + if vs_id: + vs_ids = [vs_id] + except Exception: + pass + if vs_ids: + tool_def = {"type": "file_search", "vector_store_ids": vs_ids} + if filters: + tool_def["filters"] = filters + tools.append(tool_def) + elif provider == ModelProvider.GOOGLE: + attachments = await prepare_attachments( + user=username, + target_provider=provider, + attached_ids=scoped_file_ids, + llm_config=config, + ) + elif provider == ModelProvider.CLAUDE: + attachments = await prepare_attachments( + user=username, + target_provider=provider, + attached_ids=scoped_file_ids, + llm_config=config, + ) + + attachments_per_model.append(attachments or None) + tools_per_model.append(tools or None) + + # Build judge config (if external_judge mode) + judge_config = None + if request.judge_mode == DebateJudgeMode.EXTERNAL_JUDGE and request.judge_model: + judge = request.judge_model + judge_provider = resolve_provider(judge.model_name) + judge_api_key = get_user_api_key(resolved, judge_provider.value) + judge_config = LLMConfig( + provider=judge_provider, + model_name=judge.model_name, + temperature=judge.temperature if judge.temperature is not None else request.temperature, + system_prompt=request.system_prompt, + api_key=judge_api_key, + reasoning_effort=judge.reasoning_effort if judge.reasoning_effort is not None else request.reasoning_effort, + enable_google_search=judge.enable_google_search if judge.enable_google_search is not None else request.enable_google_search, + ) + + return StreamingResponse( + debate_event_stream( + user_prompt=request.user_prompt, + context=execution_context, + member_configs=member_configs, + judge_config=judge_config, + judge_mode=request.judge_mode, + debate_format=request.debate_format, + max_rounds=request.max_rounds, + custom_format_prompt=request.custom_format_prompt, + attachments_per_model=attachments_per_model, + tools_per_model=tools_per_model, + openrouter_api_key=openrouter_key, + images=images, + ), + media_type="text/event-stream", + ) + + class TitleRequest(BaseModel): user_prompt: str response: str diff --git a/backend/app/schemas.py b/backend/app/schemas.py index 4213f15..7a657a3 100644 --- a/backend/app/schemas.py +++ b/backend/app/schemas.py @@ -76,6 +76,35 @@ class CouncilRunRequest(BaseModel): attached_file_ids: List[str] = Field(default_factory=list) scopes: List[str] = Field(default_factory=list) +class DebateJudgeMode(str, Enum): + EXTERNAL_JUDGE = "external_judge" + SELF_CONVERGENCE = "self_convergence" + DISPLAY_ONLY = "display_only" + +class DebateFormat(str, Enum): + FREE_DISCUSSION = "free_discussion" + STRUCTURED_OPPOSITION = "structured_opposition" + ITERATIVE_IMPROVEMENT = "iterative_improvement" + CUSTOM = "custom" + +class DebateRunRequest(BaseModel): + node_id: str + incoming_contexts: List[Context] = [] + user_prompt: str + debate_models: List[CouncilMemberConfig] # 2-6 models + judge_model: Optional[CouncilMemberConfig] = None + judge_mode: DebateJudgeMode = DebateJudgeMode.EXTERNAL_JUDGE + debate_format: DebateFormat = DebateFormat.FREE_DISCUSSION + custom_format_prompt: Optional[str] = None + max_rounds: int = 5 + system_prompt: Optional[str] = None + temperature: float = 0.7 + reasoning_effort: ReasoningEffort = ReasoningEffort.MEDIUM + enable_google_search: bool = False + merge_strategy: MergeStrategy = MergeStrategy.SMART + attached_file_ids: List[str] = Field(default_factory=list) + scopes: List[str] = Field(default_factory=list) + class NodeRunResponse(BaseModel): node_id: str output_context: Context diff --git a/backend/app/services/debate.py b/backend/app/services/debate.py new file mode 100644 index 0000000..d409cb9 --- /dev/null +++ b/backend/app/services/debate.py @@ -0,0 +1,371 @@ +"""Multi-round LLM Debate orchestration for ContextFlow.""" + +import asyncio +import json +import logging +from typing import AsyncGenerator, Dict, List, Any, Optional + +from app.schemas import Context, LLMConfig, DebateFormat, DebateJudgeMode +from app.services.llm import query_model_full, llm_streamer + +logger = logging.getLogger("contextflow.debate") + + +def _sse_event(data: dict) -> str: + """Format a dict as an SSE data line.""" + return f"data: {json.dumps(data)}\n\n" + + +def build_debate_prompt( + user_query: str, + debate_history: List[Dict[str, Any]], + model_name: str, + round_num: int, + debate_format: DebateFormat, + custom_prompt: Optional[str] = None, + model_index: int = 0, + total_models: int = 2, +) -> str: + """Build the prompt for a debater based on format and history.""" + history_text = "" + if debate_history: + for past_round in debate_history: + rn = past_round["round"] + history_text += f"\n--- Round {rn} ---\n" + for resp in past_round["responses"]: + history_text += f"\n[{resp['model']}]:\n{resp['response']}\n" + + if debate_format == DebateFormat.FREE_DISCUSSION: + if round_num == 1: + return ( + f"You are participating in a roundtable discussion about the following question:\n\n" + f'"{user_query}"\n\n' + f"Provide your perspective and answer to this question." + ) + return ( + f"You are participating in a roundtable discussion about the following question:\n\n" + f'"{user_query}"\n\n' + f"Here is the discussion so far:\n{history_text}\n\n" + f"This is round {round_num}. Consider what others have said, respond to their points, " + f"and refine or defend your position." + ) + + if debate_format == DebateFormat.STRUCTURED_OPPOSITION: + roles = ["FOR", "AGAINST", "DEVIL'S ADVOCATE", "MEDIATOR", "CRITIC", "SYNTHESIZER"] + role = roles[model_index % len(roles)] + if round_num == 1: + return ( + f"You are arguing {role} the following position in a structured debate:\n\n" + f'"{user_query}"\n\n' + f"Present your strongest arguments from the {role} perspective." + ) + return ( + f"You are arguing {role} the following position in a structured debate:\n\n" + f'"{user_query}"\n\n' + f"Debate history:\n{history_text}\n\n" + f"This is round {round_num}. Respond to the other participants' arguments " + f"while maintaining your {role} position. Address their strongest points." + ) + + if debate_format == DebateFormat.ITERATIVE_IMPROVEMENT: + if round_num == 1: + return ( + f"You are participating in an iterative improvement exercise on the following question:\n\n" + f'"{user_query}"\n\n' + f"Provide your best answer." + ) + return ( + f"You are participating in an iterative improvement exercise on the following question:\n\n" + f'"{user_query}"\n\n' + f"Here are the previous answers from all participants:\n{history_text}\n\n" + f"This is round {round_num}. Critique the other participants' answers, identify flaws or gaps, " + f"and provide an improved answer that incorporates the best insights from everyone." + ) + + if debate_format == DebateFormat.CUSTOM and custom_prompt: + prompt = custom_prompt + prompt = prompt.replace("{history}", history_text or "(No history yet)") + prompt = prompt.replace("{round}", str(round_num)) + prompt = prompt.replace("{model_name}", model_name) + prompt = prompt.replace("{question}", user_query) + return prompt + + # Fallback to free discussion + if round_num == 1: + return f'Provide your answer to the following question:\n\n"{user_query}"' + return ( + f'Question: "{user_query}"\n\n' + f"Previous discussion:\n{history_text}\n\n" + f"Round {round_num}: Provide your updated response." + ) + + +async def debate_round( + configs: List[LLMConfig], + context: Context, + user_prompt: str, + debate_history: List[Dict[str, Any]], + round_num: int, + debate_format: DebateFormat, + custom_prompt: Optional[str] = None, + attachments_per_model: Optional[List[Optional[List[Dict[str, Any]]]]] = None, + tools_per_model: Optional[List[Optional[List[Dict[str, Any]]]]] = None, + openrouter_api_key: Optional[str] = None, + images: Optional[List[Dict[str, Any]]] = None, +) -> AsyncGenerator[Dict[str, Any], None]: + """Query all debate models in parallel for one round, yielding as each completes.""" + + async def _query_one(idx: int, config: LLMConfig) -> Dict[str, Any]: + prompt = build_debate_prompt( + user_prompt, debate_history, config.model_name, + round_num, debate_format, custom_prompt, + model_index=idx, total_models=len(configs), + ) + atts = attachments_per_model[idx] if attachments_per_model else None + tls = tools_per_model[idx] if tools_per_model else None + try: + response = await query_model_full( + context, prompt, config, + attachments=atts, tools=tls, + openrouter_api_key=openrouter_api_key, + images=images if round_num == 1 else None, # Only send images in round 1 + ) + return {"model": config.model_name, "response": response} + except Exception as e: + logger.error("Debate round %d failed for %s: %s", round_num, config.model_name, e) + return {"model": config.model_name, "response": f"[Error: {e}]"} + + tasks = { + asyncio.ensure_future(_query_one(i, cfg)): i + for i, cfg in enumerate(configs) + } + for coro in asyncio.as_completed(tasks.keys()): + result = await coro + yield result + + +async def judge_evaluate_round( + judge_config: LLMConfig, + debate_history: List[Dict[str, Any]], + user_query: str, + openrouter_api_key: Optional[str] = None, +) -> Dict[str, Any]: + """Judge decides if debate should continue after a round.""" + last_round = len(debate_history) + history_text = "" + for past_round in debate_history: + rn = past_round["round"] + history_text += f"\n--- Round {rn} ---\n" + for resp in past_round["responses"]: + history_text += f"\n[{resp['model']}]:\n{resp['response']}\n" + + prompt = ( + f"You are the judge of a multi-model debate on the following question:\n" + f'"{user_query}"\n\n' + f"Debate history (Round 1 to {last_round}):\n{history_text}\n\n" + f"Evaluate whether the debate has reached a satisfactory conclusion.\n" + f"Consider: Have the key points been thoroughly explored? Is there consensus?\n" + f"Are there unresolved disagreements worth continuing?\n\n" + f"Respond with exactly one of:\n" + f"CONTINUE - if the debate should go on (explain why briefly)\n" + f"STOP - if a clear conclusion has been reached (explain why briefly)" + ) + + empty_context = Context(messages=[]) + try: + response = await query_model_full( + empty_context, prompt, judge_config, + openrouter_api_key=openrouter_api_key, + ) + should_continue = "CONTINUE" in response.upper().split("\n")[0] + return {"continue": should_continue, "reasoning": response} + except Exception as e: + logger.error("Judge evaluation failed: %s", e) + return {"continue": False, "reasoning": f"[Judge error: {e}]"} + + +async def check_self_convergence( + configs: List[LLMConfig], + round_responses: List[Dict[str, Any]], + openrouter_api_key: Optional[str] = None, +) -> Dict[str, Any]: + """Check if debate responses have converged using the first available model.""" + responses_text = "\n\n".join( + f"[{r['model']}]:\n{r['response']}" for r in round_responses + ) + prompt = ( + f"Below are the responses from the latest round of a debate:\n\n" + f"{responses_text}\n\n" + f"Do all participants essentially agree on the answer? Respond ONLY with:\n" + f"CONVERGED - if there is clear consensus\n" + f"DIVERGENT - if there are still significant disagreements" + ) + + empty_context = Context(messages=[]) + # Use the first config as the convergence checker + check_config = configs[0] + try: + response = await query_model_full( + empty_context, prompt, check_config, + openrouter_api_key=openrouter_api_key, + ) + converged = "CONVERGED" in response.upper().split("\n")[0] + return {"converged": converged, "reasoning": response} + except Exception as e: + logger.error("Convergence check failed: %s", e) + return {"converged": False, "reasoning": f"[Convergence check error: {e}]"} + + +async def judge_final_verdict( + judge_config: LLMConfig, + debate_history: List[Dict[str, Any]], + user_query: str, + openrouter_api_key: Optional[str] = None, +) -> AsyncGenerator[str, None]: + """Stream the judge's final verdict/synthesis.""" + history_text = "" + for past_round in debate_history: + rn = past_round["round"] + history_text += f"\n--- Round {rn} ---\n" + for resp in past_round["responses"]: + history_text += f"\n[{resp['model']}]:\n{resp['response']}\n" + + prompt = ( + f"You are the judge of a multi-model debate. Below is the full debate transcript.\n\n" + f'Question: "{user_query}"\n\n' + f"{history_text}\n\n" + f"As the judge, provide:\n" + f"1. A summary of the key arguments from each participant\n" + f"2. An evaluation of the strengths and weaknesses of each position\n" + f"3. Your final verdict: the best, most accurate, and most comprehensive answer " + f"to the original question, synthesizing the best insights from the debate." + ) + + empty_context = Context(messages=[]) + async for chunk in llm_streamer( + empty_context, prompt, judge_config, + openrouter_api_key=openrouter_api_key, + ): + yield chunk + + +async def debate_event_stream( + user_prompt: str, + context: Context, + member_configs: List[LLMConfig], + judge_config: Optional[LLMConfig], + judge_mode: DebateJudgeMode, + debate_format: DebateFormat, + max_rounds: int = 5, + custom_format_prompt: Optional[str] = None, + attachments_per_model: Optional[List[Optional[List[Dict[str, Any]]]]] = None, + tools_per_model: Optional[List[Optional[List[Dict[str, Any]]]]] = None, + openrouter_api_key: Optional[str] = None, + images: Optional[List[Dict[str, Any]]] = None, +) -> AsyncGenerator[str, None]: + """Master orchestrator yielding SSE JSON events through the debate process.""" + + model_names = [c.model_name for c in member_configs] + yield _sse_event({ + "type": "debate_start", + "data": { + "max_rounds": max_rounds, + "format": debate_format.value, + "judge_mode": judge_mode.value, + "models": model_names, + }, + }) + + debate_history: List[Dict[str, Any]] = [] + + for round_num in range(1, max_rounds + 1): + yield _sse_event({"type": "round_start", "data": {"round": round_num}}) + + round_responses: List[Dict[str, Any]] = [] + async for result in debate_round( + member_configs, context, user_prompt, + debate_history, round_num, debate_format, custom_format_prompt, + attachments_per_model=attachments_per_model, + tools_per_model=tools_per_model, + openrouter_api_key=openrouter_api_key, + images=images, + ): + round_responses.append(result) + yield _sse_event({ + "type": "round_model_complete", + "data": {"round": round_num, "model": result["model"], "response": result["response"]}, + }) + + debate_history.append({"round": round_num, "responses": round_responses}) + + yield _sse_event({ + "type": "round_complete", + "data": {"round": round_num, "responses": round_responses}, + }) + + if not round_responses: + yield _sse_event({ + "type": "error", + "data": {"message": "All debate models failed to respond."}, + }) + return + + # Check stop condition (skip on last round) + if round_num < max_rounds: + if judge_mode == DebateJudgeMode.EXTERNAL_JUDGE and judge_config: + decision = await judge_evaluate_round( + judge_config, debate_history, user_prompt, + openrouter_api_key=openrouter_api_key, + ) + yield _sse_event({ + "type": "judge_decision", + "data": {"round": round_num, **decision}, + }) + if not decision["continue"]: + break + + elif judge_mode == DebateJudgeMode.SELF_CONVERGENCE: + convergence = await check_self_convergence( + member_configs, round_responses, + openrouter_api_key=openrouter_api_key, + ) + yield _sse_event({ + "type": "convergence_check", + "data": {"round": round_num, **convergence}, + }) + if convergence["converged"]: + break + # DISPLAY_ONLY: just continue to next round + + # Final synthesis + if judge_mode == DebateJudgeMode.EXTERNAL_JUDGE and judge_config: + yield _sse_event({ + "type": "final_start", + "data": {"model": judge_config.model_name}, + }) + + full_verdict = "" + async for chunk in judge_final_verdict( + judge_config, debate_history, user_prompt, + openrouter_api_key=openrouter_api_key, + ): + full_verdict += chunk + yield _sse_event({"type": "final_chunk", "data": {"chunk": chunk}}) + + yield _sse_event({ + "type": "final_complete", + "data": {"model": judge_config.model_name, "response": full_verdict}, + }) + + elif judge_mode == DebateJudgeMode.SELF_CONVERGENCE: + # Use the last round's responses as the final answer + last_responses = debate_history[-1]["responses"] if debate_history else [] + # Pick the longest response as the "best" convergent answer + if last_responses: + best = max(last_responses, key=lambda r: len(r.get("response", ""))) + yield _sse_event({ + "type": "final_complete", + "data": {"model": best["model"], "response": best["response"]}, + }) + + yield _sse_event({"type": "debate_complete"}) diff --git a/frontend/src/components/Sidebar.tsx b/frontend/src/components/Sidebar.tsx index 64cd79a..aeb164b 100644 --- a/frontend/src/components/Sidebar.tsx +++ b/frontend/src/components/Sidebar.tsx @@ -2,7 +2,7 @@ import React, { useState, useEffect, useRef, useMemo } from 'react'; import { useReactFlow } from 'reactflow'; import useFlowStore from '../store/flowStore'; import { useAuthStore } from '../store/authStore'; -import type { NodeData, Trace, Message, MergedTrace, MergeStrategy, CouncilData, CouncilMemberConfig } from '../store/flowStore'; +import type { NodeData, Trace, Message, MergedTrace, MergeStrategy, CouncilData, CouncilMemberConfig, DebateData, DebateRound } from '../store/flowStore'; import type { Edge } from 'reactflow'; import ReactMarkdown from 'react-markdown'; import remarkGfm from 'remark-gfm'; @@ -17,7 +17,7 @@ const preprocessLaTeX = (content: string): string => { .replace(/\\\(([\s\S]*?)\\\)/g, (_, math) => `$${math}$`); }; -import { Play, Settings, Info, ChevronLeft, ChevronRight, ChevronDown, Maximize2, Edit3, X, Check, FileText, MessageCircle, Send, GripVertical, GitMerge, Trash2, AlertCircle, Loader2, Navigation, Upload, Search, Link, Layers, Eye, EyeOff, Copy, ClipboardCheck, Users } from 'lucide-react'; +import { Play, Settings, Info, ChevronLeft, ChevronRight, ChevronDown, Maximize2, Edit3, X, Check, FileText, MessageCircle, Send, GripVertical, GitMerge, Trash2, AlertCircle, Loader2, Navigation, Upload, Search, Link, Layers, Eye, EyeOff, Copy, ClipboardCheck, Users, MessageSquare } from 'lucide-react'; interface SidebarProps { isOpen: boolean; @@ -109,6 +109,13 @@ const Sidebar: React.FC<SidebarProps> = ({ isOpen, onToggle, onInteract }) => { const [quickChatCouncilData, setQuickChatCouncilData] = useState<CouncilData | null>(null); const [quickChatCouncilConfigOpen, setQuickChatCouncilConfigOpen] = useState(false); + // Debate mode states + const [debateStage, setDebateStage] = useState<string>(''); + const [debateStreamBuffer, setDebateStreamBuffer] = useState(''); + const [debateTab, setDebateTab] = useState<'final' | 'timeline' | 'per-model'>('timeline'); + const [debateTimelineExpandedRounds, setDebateTimelineExpandedRounds] = useState<Set<number>>(new Set()); + const [debatePerModelSelected, setDebatePerModelSelected] = useState<string>(''); + const selectedNode = nodes.find((n) => n.id === selectedNodeId); // Reset stream buffer and modal states when node changes @@ -640,6 +647,263 @@ const Sidebar: React.FC<SidebarProps> = ({ isOpen, onToggle, onInteract }) => { } }; + // Debate mode: multi-round LLM debate execution + const handleRunDebate = async () => { + if (!selectedNode) return; + const debateModels: CouncilMemberConfig[] = selectedNode.data.debateModels || []; + if (debateModels.length < 2) return; + + const tracesCheck = checkActiveTracesComplete(); + if (!tracesCheck.complete) return; + + const runningNodeId = selectedNode.id; + const runningPrompt = selectedNode.data.userPrompt; + const querySentAt = Date.now(); + const judgeMode = selectedNode.data.debateJudgeMode || 'external_judge'; + const debateFormat = selectedNode.data.debateFormat || 'free_discussion'; + const maxRounds = selectedNode.data.debateMaxRounds || 5; + + updateNodeData(runningNodeId, { + status: 'loading', + response: '', + querySentAt, + debateData: { + rounds: [], + finalVerdict: null, + config: { judgeMode, format: debateFormat, maxRounds }, + }, + }); + setStreamBuffer(''); + setDebateStreamBuffer(''); + setDebateStage('Starting debate...'); + setStreamingNodeId(runningNodeId); + + const context = getActiveContext(runningNodeId); + const projectPath = currentBlueprintPath || 'untitled'; + const traceNodeIds = new Set<string>(); + traceNodeIds.add(runningNodeId); + const visited = new Set<string>(); + const queue = [runningNodeId]; + while (queue.length > 0) { + const currentNodeId = queue.shift()!; + if (visited.has(currentNodeId)) continue; + visited.add(currentNodeId); + const incomingEdges = edges.filter(e => e.target === currentNodeId); + for (const edge of incomingEdges) { + if (!visited.has(edge.source)) { + traceNodeIds.add(edge.source); + queue.push(edge.source); + } + } + } + const scopes = Array.from(traceNodeIds).map(nodeId => `${projectPath}/${nodeId}`); + + const attachedFiles = selectedNode.data.attachedFileIds || []; + const effectivePrompt = runningPrompt?.trim() + ? runningPrompt + : attachedFiles.length > 0 + ? 'Please analyze the attached files.' + : ''; + + try { + const judgeModelConfig = selectedNode.data.judgeModel || debateModels[0]; + const response = await fetch(`/api/run_debate_stream?user=${encodeURIComponent(user?.username || 'test')}`, { + method: 'POST', + headers: { 'Content-Type': 'application/json', ...getAuthHeader() }, + body: JSON.stringify({ + node_id: runningNodeId, + incoming_contexts: [{ messages: context }], + user_prompt: effectivePrompt, + debate_models: debateModels.map((cfg) => ({ + model_name: cfg.model, + temperature: cfg.temperature ?? null, + reasoning_effort: cfg.reasoningEffort ?? null, + enable_google_search: cfg.enableWebSearch ?? null, + })), + judge_model: judgeMode === 'external_judge' ? { + model_name: judgeModelConfig.model, + temperature: judgeModelConfig.temperature ?? null, + reasoning_effort: judgeModelConfig.reasoningEffort ?? null, + enable_google_search: judgeModelConfig.enableWebSearch ?? null, + } : null, + judge_mode: judgeMode, + debate_format: debateFormat, + custom_format_prompt: selectedNode.data.debateCustomPrompt || null, + max_rounds: maxRounds, + system_prompt: selectedNode.data.systemPrompt || null, + temperature: selectedNode.data.temperature, + reasoning_effort: selectedNode.data.reasoningEffort || 'medium', + enable_google_search: selectedNode.data.enableGoogleSearch !== false, + merge_strategy: selectedNode.data.mergeStrategy || 'smart', + attached_file_ids: attachedFiles, + scopes, + }), + }); + + if (!response.body) return; + const reader = response.body.getReader(); + const decoder = new TextDecoder(); + let sseBuffer = ''; + const debateRounds: DebateRound[] = []; + let currentRound = 0; + let currentRoundResponses: Array<{ model: string; response: string }> = []; + let finalModel = ''; + let finalFull = ''; + + while (true) { + const { value, done } = await reader.read(); + if (done) break; + sseBuffer += decoder.decode(value, { stream: true }); + + const parts = sseBuffer.split('\n\n'); + sseBuffer = parts.pop() || ''; + + for (const part of parts) { + const line = part.trim(); + if (!line.startsWith('data: ')) continue; + let evt: any; + try { + evt = JSON.parse(line.slice(6)); + } catch { continue; } + + switch (evt.type) { + case 'debate_start': + setDebateStage(`Debate started (${evt.data.models.length} models, max ${evt.data.max_rounds} rounds)`); + break; + case 'round_start': + currentRound = evt.data.round; + currentRoundResponses = []; + setDebateStage(`Round ${currentRound}/${maxRounds}: Collecting responses...`); + break; + case 'round_model_complete': + currentRoundResponses = [...currentRoundResponses, { model: evt.data.model, response: evt.data.response }]; + setDebateStage(`Round ${currentRound}/${maxRounds}: ${currentRoundResponses.length}/${debateModels.length} models done`); + break; + case 'round_complete': { + const roundData: DebateRound = { round: evt.data.round, responses: evt.data.responses }; + debateRounds.push(roundData); + updateNodeData(runningNodeId, { + debateData: { + rounds: [...debateRounds], + finalVerdict: null, + config: { judgeMode, format: debateFormat, maxRounds }, + }, + }); + break; + } + case 'judge_decision': { + const lastRound = debateRounds[debateRounds.length - 1]; + if (lastRound) { + lastRound.judgeDecision = { continue: evt.data.continue, reasoning: evt.data.reasoning }; + updateNodeData(runningNodeId, { + debateData: { + rounds: [...debateRounds], + finalVerdict: null, + config: { judgeMode, format: debateFormat, maxRounds }, + }, + }); + } + if (!evt.data.continue) { + setDebateStage('Judge stopped debate. Generating final verdict...'); + } else { + setDebateStage(`Judge: Continue to round ${currentRound + 1}...`); + } + break; + } + case 'convergence_check': { + const lastRound2 = debateRounds[debateRounds.length - 1]; + if (lastRound2) { + lastRound2.converged = evt.data.converged; + updateNodeData(runningNodeId, { + debateData: { + rounds: [...debateRounds], + finalVerdict: null, + config: { judgeMode, format: debateFormat, maxRounds }, + }, + }); + } + if (evt.data.converged) { + setDebateStage('Consensus reached!'); + } + break; + } + case 'final_start': + finalModel = evt.data.model; + setDebateStage('Judge synthesizing final verdict...'); + setDebateStreamBuffer(''); + break; + case 'final_chunk': + finalFull += evt.data.chunk; + setDebateStreamBuffer(finalFull); + setStreamBuffer(finalFull); + break; + case 'final_complete': { + finalModel = evt.data.model; + finalFull = evt.data.response; + const responseReceivedAt = Date.now(); + const debateData: DebateData = { + rounds: debateRounds, + finalVerdict: { model: finalModel, response: finalFull }, + config: { judgeMode, format: debateFormat, maxRounds }, + }; + const newUserMsg = { id: `msg_${Date.now()}_u`, role: 'user', content: runningPrompt }; + const newAssistantMsg = { id: `msg_${Date.now()}_a`, role: 'assistant', content: finalFull }; + updateNodeData(runningNodeId, { + status: 'success', + response: finalFull, + responseReceivedAt, + debateData, + messages: [...context, newUserMsg, newAssistantMsg] as any, + }); + setDebateStage(''); + generateTitle(runningNodeId, runningPrompt, finalFull); + break; + } + case 'debate_complete': { + // If no final verdict (display_only or self_convergence without explicit final_complete) + const currentNode = nodes.find(n => n.id === runningNodeId); + if (currentNode?.data.status === 'loading') { + const responseReceivedAt = Date.now(); + const lastRoundResp = debateRounds.length > 0 ? debateRounds[debateRounds.length - 1].responses : []; + const bestResponse = lastRoundResp.length > 0 + ? lastRoundResp.reduce((a, b) => a.response.length > b.response.length ? a : b).response + : ''; + const debateData: DebateData = { + rounds: debateRounds, + finalVerdict: finalFull ? { model: finalModel, response: finalFull } : null, + config: { judgeMode, format: debateFormat, maxRounds }, + }; + const displayResponse = finalFull || bestResponse; + const newUserMsg = { id: `msg_${Date.now()}_u`, role: 'user', content: runningPrompt }; + const newAssistantMsg = { id: `msg_${Date.now()}_a`, role: 'assistant', content: displayResponse }; + updateNodeData(runningNodeId, { + status: 'success', + response: displayResponse, + responseReceivedAt, + debateData, + messages: [...context, newUserMsg, newAssistantMsg] as any, + }); + setDebateStage(''); + if (displayResponse) generateTitle(runningNodeId, runningPrompt, displayResponse); + } + break; + } + case 'error': + updateNodeData(runningNodeId, { status: 'error' }); + setDebateStage(''); + break; + } + } + } + } catch (error) { + console.error(error); + updateNodeData(runningNodeId, { status: 'error' }); + setDebateStage(''); + } finally { + setStreamingNodeId(prev => prev === runningNodeId ? null : prev); + } + }; + const handleChange = (field: keyof NodeData, value: any) => { updateNodeData(selectedNode.id, { [field]: value }); }; @@ -1844,20 +2108,218 @@ const Sidebar: React.FC<SidebarProps> = ({ isOpen, onToggle, onInteract }) => { <div> <div className="flex items-center justify-between mb-1"> <label className="block text-sm font-medium text-gray-700">Model</label> - <button - onClick={() => handleChange('councilMode', !selectedNode.data.councilMode)} - className={`flex items-center gap-1 text-xs px-2 py-0.5 rounded-full transition-colors ${ - selectedNode.data.councilMode - ? 'bg-amber-100 text-amber-700 border border-amber-300 dark:bg-amber-900/50 dark:text-amber-300 dark:border-amber-700' - : isDark ? 'bg-gray-700 text-gray-400 hover:bg-gray-600' : 'bg-gray-100 text-gray-500 hover:bg-gray-200' - }`} - > - <Users size={11} /> - Council {selectedNode.data.councilMode ? 'ON' : 'OFF'} - </button> + <div className="flex gap-1"> + <button + onClick={() => { + const next = !selectedNode.data.councilMode; + handleChange('councilMode', next); + if (next) handleChange('debateMode', false); + }} + className={`flex items-center gap-1 text-xs px-2 py-0.5 rounded-full transition-colors ${ + selectedNode.data.councilMode + ? 'bg-amber-100 text-amber-700 border border-amber-300 dark:bg-amber-900/50 dark:text-amber-300 dark:border-amber-700' + : isDark ? 'bg-gray-700 text-gray-400 hover:bg-gray-600' : 'bg-gray-100 text-gray-500 hover:bg-gray-200' + }`} + > + <Users size={11} /> + Council {selectedNode.data.councilMode ? 'ON' : 'OFF'} + </button> + <button + onClick={() => { + const next = !selectedNode.data.debateMode; + handleChange('debateMode', next); + if (next) handleChange('councilMode', false); + }} + className={`flex items-center gap-1 text-xs px-2 py-0.5 rounded-full transition-colors ${ + selectedNode.data.debateMode + ? 'bg-cyan-100 text-cyan-700 border border-cyan-300 dark:bg-cyan-900/50 dark:text-cyan-300 dark:border-cyan-700' + : isDark ? 'bg-gray-700 text-gray-400 hover:bg-gray-600' : 'bg-gray-100 text-gray-500 hover:bg-gray-200' + }`} + > + <MessageSquare size={11} /> + Debate {selectedNode.data.debateMode ? 'ON' : 'OFF'} + </button> + </div> </div> - {!selectedNode.data.councilMode ? ( + {selectedNode.data.debateMode ? ( + /* Debate mode: multi-model selector + judge config */ + <div className={`space-y-2 p-2 rounded border ${isDark ? 'bg-gray-900 border-cyan-800/50' : 'bg-cyan-50/50 border-cyan-200'}`}> + <div> + <label className={`block text-xs font-semibold mb-1 ${isDark ? 'text-gray-400' : 'text-gray-600'}`}>Debate Participants (pick 2-6)</label> + <div className="grid grid-cols-1 gap-0.5 max-h-40 overflow-y-auto"> + {(() => { + const debateModelsList = [ + { value: 'claude-sonnet-4-5', label: 'claude-sonnet-4.5' }, + { value: 'claude-opus-4', label: 'claude-opus-4' }, + { value: 'claude-opus-4-5', label: 'claude-opus-4.5' }, + { value: 'claude-opus-4-6', label: 'claude-opus-4.6' }, + { value: 'gemini-2.5-flash', label: 'gemini-2.5-flash' }, + { value: 'gemini-2.5-flash-lite', label: 'gemini-2.5-flash-lite' }, + { value: 'gemini-3-pro-preview', label: 'gemini-3-pro-preview' }, + { value: 'gpt-4.1', label: 'gpt-4.1' }, + { value: 'gpt-4o', label: 'gpt-4o' }, + { value: 'gpt-5', label: 'gpt-5' }, + { value: 'gpt-5-mini', label: 'gpt-5-mini' }, + { value: 'gpt-5-nano', label: 'gpt-5-nano' }, + { value: 'gpt-5.1', label: 'gpt-5.1' }, + { value: 'gpt-5.2', label: 'gpt-5.2' }, + { value: 'o3', label: 'o3', premium: true }, + ]; + const members: CouncilMemberConfig[] = selectedNode.data.debateModels || []; + const isReasoningModel = (v: string) => ['gpt-5', 'gpt-5-mini', 'gpt-5-nano', 'gpt-5-pro', 'gpt-5.1', 'gpt-5.2', 'gpt-5.2-pro', 'o3'].includes(v); + const updateDebateMember = (modelName: string, field: string, value: any) => { + const updated = [...members]; + const idx = updated.findIndex(c => c.model === modelName); + if (idx >= 0) { + updated[idx] = { ...updated[idx], [field]: value }; + handleChange('debateModels', updated); + } + }; + return debateModelsList.map(m => { + const selected = members.some(c => c.model === m.value); + const disabled = (m as any).premium && !canUsePremiumModels; + const cfg = members.find(c => c.model === m.value); + return ( + <div key={m.value}> + <label className={`flex items-center gap-1.5 text-xs py-0.5 px-1 rounded cursor-pointer ${ + disabled ? 'opacity-40 cursor-not-allowed' : selected + ? isDark ? 'bg-cyan-900/40 text-cyan-200' : 'bg-cyan-100 text-cyan-800' + : isDark ? 'hover:bg-gray-800 text-gray-300' : 'hover:bg-gray-100 text-gray-700' + }`}> + <input + type="checkbox" + checked={selected} + disabled={disabled} + onChange={() => { + const next = selected + ? members.filter(c => c.model !== m.value) + : [...members, { model: m.value }]; + handleChange('debateModels', next); + }} + className="w-3 h-3 accent-cyan-500" + /> + {m.label} + </label> + {selected && cfg && ( + <div className={`flex items-center gap-2 ml-5 mt-0.5 mb-1 px-1 py-0.5 text-xs rounded ${isDark ? 'bg-gray-800/60' : 'bg-gray-100/80'}`}> + <label className={`flex items-center gap-1 ${isDark ? 'text-gray-400' : 'text-gray-500'}`}> + T: + <input + type="number" + min={0} max={2} step={0.1} + value={cfg.temperature ?? selectedNode.data.temperature} + disabled={isReasoningModel(m.value)} + onChange={(e) => updateDebateMember(m.value, 'temperature', parseFloat(e.target.value) || 0)} + className={`w-12 border rounded px-1 py-0 text-xs ${isReasoningModel(m.value) ? 'opacity-50 cursor-not-allowed' : ''} ${isDark ? 'bg-gray-700 border-gray-600 text-gray-200' : 'bg-white border-gray-300'}`} + /> + </label> + {isReasoningModel(m.value) && ( + <label className={`flex items-center gap-1 ${isDark ? 'text-gray-400' : 'text-gray-500'}`}> + Effort: + <select + value={cfg.reasoningEffort ?? selectedNode.data.reasoningEffort ?? 'medium'} + onChange={(e) => updateDebateMember(m.value, 'reasoningEffort', e.target.value)} + className={`border rounded px-1 py-0 text-xs ${isDark ? 'bg-gray-700 border-gray-600 text-gray-200' : 'bg-white border-gray-300'}`} + > + <option value="low">Low</option> + <option value="medium">Med</option> + <option value="high">High</option> + </select> + </label> + )} + </div> + )} + </div> + ); + }); + })()} + </div> + </div> + + {/* Judge Mode */} + <div> + <label className={`block text-xs font-semibold mb-1 ${isDark ? 'text-gray-400' : 'text-gray-600'}`}>Judge Mode</label> + <select + value={selectedNode.data.debateJudgeMode || 'external_judge'} + onChange={(e) => handleChange('debateJudgeMode', e.target.value)} + className={`w-full border rounded-md p-1.5 text-xs ${isDark ? 'bg-gray-800 border-gray-600 text-gray-200' : 'bg-white border-gray-300'}`} + > + <option value="external_judge">External Judge</option> + <option value="self_convergence">Self-Convergence</option> + <option value="display_only">Display Only</option> + </select> + </div> + + {/* Judge Model (only for external_judge) */} + {(selectedNode.data.debateJudgeMode || 'external_judge') === 'external_judge' && ( + <div> + <label className={`block text-xs font-semibold mb-1 ${isDark ? 'text-gray-400' : 'text-gray-600'}`}>Judge Model</label> + <select + value={selectedNode.data.judgeModel?.model || (selectedNode.data.debateModels || [])[0]?.model || ''} + onChange={(e) => { + const existing = (selectedNode.data.debateModels || []).find((c: CouncilMemberConfig) => c.model === e.target.value); + handleChange('judgeModel', existing ? { ...existing } : { model: e.target.value }); + }} + className={`w-full border rounded-md p-1.5 text-xs ${isDark ? 'bg-gray-800 border-gray-600 text-gray-200' : 'bg-white border-gray-300'}`} + > + {[ + { value: 'claude-sonnet-4-5', label: 'claude-sonnet-4.5' }, + { value: 'claude-opus-4-6', label: 'claude-opus-4.6' }, + { value: 'gemini-2.5-flash', label: 'gemini-2.5-flash' }, + { value: 'gemini-3-pro-preview', label: 'gemini-3-pro-preview' }, + { value: 'gpt-5', label: 'gpt-5' }, + { value: 'gpt-5.1', label: 'gpt-5.1' }, + { value: 'gpt-5.2', label: 'gpt-5.2' }, + ].map(m => ( + <option key={m.value} value={m.value}>{m.label}</option> + ))} + </select> + </div> + )} + + {/* Debate Format */} + <div> + <label className={`block text-xs font-semibold mb-1 ${isDark ? 'text-gray-400' : 'text-gray-600'}`}>Debate Format</label> + <select + value={selectedNode.data.debateFormat || 'free_discussion'} + onChange={(e) => handleChange('debateFormat', e.target.value)} + className={`w-full border rounded-md p-1.5 text-xs ${isDark ? 'bg-gray-800 border-gray-600 text-gray-200' : 'bg-white border-gray-300'}`} + > + <option value="free_discussion">Free Discussion</option> + <option value="structured_opposition">Structured Opposition</option> + <option value="iterative_improvement">Iterative Improvement</option> + <option value="custom">Custom Prompt</option> + </select> + </div> + + {/* Custom Prompt (only for custom format) */} + {selectedNode.data.debateFormat === 'custom' && ( + <div> + <label className={`block text-xs font-semibold mb-1 ${isDark ? 'text-gray-400' : 'text-gray-600'}`}>Custom Prompt</label> + <textarea + value={selectedNode.data.debateCustomPrompt || ''} + onChange={(e) => handleChange('debateCustomPrompt', e.target.value)} + placeholder="Use {history}, {round}, {model_name}, {question} as placeholders" + rows={3} + className={`w-full border rounded-md p-1.5 text-xs resize-y ${isDark ? 'bg-gray-800 border-gray-600 text-gray-200 placeholder-gray-500' : 'bg-white border-gray-300 placeholder-gray-400'}`} + /> + </div> + )} + + {/* Max Rounds */} + <div> + <label className={`block text-xs font-semibold mb-1 ${isDark ? 'text-gray-400' : 'text-gray-600'}`}>Max Rounds</label> + <input + type="number" + min={1} max={10} + value={selectedNode.data.debateMaxRounds || 5} + onChange={(e) => handleChange('debateMaxRounds', Math.max(1, Math.min(10, parseInt(e.target.value) || 5)))} + className={`w-20 border rounded-md p-1.5 text-xs ${isDark ? 'bg-gray-800 border-gray-600 text-gray-200' : 'bg-white border-gray-300'}`} + /> + </div> + </div> + ) : !selectedNode.data.councilMode ? ( /* Single model selector */ <select value={selectedNode.data.model} @@ -2500,6 +2962,19 @@ const Sidebar: React.FC<SidebarProps> = ({ isOpen, onToggle, onInteract }) => { Inspired by <a href="https://github.com/karpathy/llm-council" target="_blank" rel="noopener noreferrer" className={`underline decoration-dotted ${isDark ? 'text-gray-500 hover:text-gray-400' : 'text-gray-500 hover:text-gray-600'}`}>karpathy/llm-council</a> </div> </> + ) : selectedNode.data.debateMode ? ( + <button + onClick={handleRunDebate} + disabled={selectedNode.data.status === 'loading' || !activeTracesCheck.complete || (selectedNode.data.debateModels || []).length < 2} + className={`w-full py-2 px-4 rounded-md flex items-center justify-center gap-2 transition-colors ${ + selectedNode.data.status === 'loading' || !activeTracesCheck.complete || (selectedNode.data.debateModels || []).length < 2 + ? 'bg-gray-300 text-gray-500 cursor-not-allowed dark:bg-gray-600 dark:text-gray-400' + : 'bg-cyan-600 text-white hover:bg-cyan-700' + }`} + > + {selectedNode.data.status === 'loading' ? <Loader2 className="animate-spin" size={16} /> : <MessageSquare size={16} />} + {selectedNode.data.status === 'loading' && debateStage ? debateStage : `Run Debate (${(selectedNode.data.debateModels || []).length})`} + </button> ) : ( <button onClick={handleRun} @@ -2713,6 +3188,198 @@ const Sidebar: React.FC<SidebarProps> = ({ isOpen, onToggle, onInteract }) => { )} </div> + ) : selectedNode.data.debateMode && selectedNode.data.debateData && (selectedNode.data.debateData.rounds.length > 0 || selectedNode.data.status === 'loading') ? ( + <div> + {/* Debate tab bar */} + <div className={`flex gap-0.5 mb-2 text-xs border-b ${isDark ? 'border-gray-700' : 'border-gray-200'}`}> + {(selectedNode.data.debateData.config.judgeMode !== 'display_only' + ? (['final', 'timeline', 'per-model'] as const) + : (['timeline', 'per-model'] as const) + ).map(tab => ( + <button + key={tab} + onClick={() => setDebateTab(tab)} + className={`px-3 py-1.5 rounded-t transition-colors capitalize ${ + debateTab === tab + ? isDark ? 'bg-gray-800 text-cyan-300 border-b-2 border-cyan-400' : 'bg-white text-cyan-700 border-b-2 border-cyan-500' + : isDark ? 'text-gray-500 hover:text-gray-300' : 'text-gray-400 hover:text-gray-600' + }`} + > + {tab === 'final' ? 'Final Answer' : tab === 'timeline' ? 'Timeline' : 'Per-Model'} + </button> + ))} + </div> + + {/* Final Answer tab */} + {debateTab === 'final' && ( + <div> + {selectedNode.data.debateData.finalVerdict ? ( + <> + <div className={`text-xs mb-1 ${isDark ? 'text-gray-500' : 'text-gray-400'}`}>Judge: {selectedNode.data.debateData.finalVerdict.model}</div> + <div className={`p-3 rounded-md border min-h-[150px] text-sm prose prose-sm max-w-none ${ + isDark ? 'bg-gray-900 border-gray-700 prose-invert text-gray-200' : 'bg-gray-50 border-gray-200 text-gray-900' + }`}> + {rawTextMode ? ( + <pre className="whitespace-pre-wrap break-words">{selectedNode.data.debateData.finalVerdict.response}</pre> + ) : ( + <ReactMarkdown remarkPlugins={[remarkGfm, remarkMath]} rehypePlugins={[rehypeKatex]}> + {preprocessLaTeX(selectedNode.data.debateData.finalVerdict.response)} + </ReactMarkdown> + )} + </div> + </> + ) : selectedNode.data.status === 'loading' ? ( + <div className={`p-3 rounded-md border min-h-[150px] text-sm ${ + isDark ? 'bg-gray-900 border-gray-700' : 'bg-gray-50 border-gray-200' + }`}> + {debateStreamBuffer ? ( + rawTextMode ? ( + <pre className="whitespace-pre-wrap break-words">{debateStreamBuffer}</pre> + ) : ( + <ReactMarkdown remarkPlugins={[remarkGfm, remarkMath]} rehypePlugins={[rehypeKatex]}> + {preprocessLaTeX(debateStreamBuffer)} + </ReactMarkdown> + ) + ) : ( + <div className={`flex items-center gap-2 ${isDark ? 'text-cyan-400' : 'text-cyan-600'}`}> + <Loader2 className="animate-spin" size={14} /> Waiting for final verdict... + </div> + )} + </div> + ) : ( + <div className={`p-3 rounded border min-h-[100px] flex items-center justify-center text-sm ${isDark ? 'bg-gray-900 border-gray-700 text-gray-500' : 'bg-gray-50 border-gray-200 text-gray-400'}`}> + No final verdict + </div> + )} + </div> + )} + + {/* Debate Timeline tab */} + {debateTab === 'timeline' && ( + <div className="space-y-2 max-h-[500px] overflow-y-auto"> + {selectedNode.data.debateData.rounds.map((round) => { + const isExpanded = debateTimelineExpandedRounds.has(round.round); + return ( + <div key={round.round} className={`rounded border ${isDark ? 'border-gray-700' : 'border-gray-200'}`}> + <button + onClick={() => { + const next = new Set(debateTimelineExpandedRounds); + if (isExpanded) next.delete(round.round); else next.add(round.round); + setDebateTimelineExpandedRounds(next); + }} + className={`w-full flex items-center justify-between px-3 py-2 text-xs font-semibold ${ + isDark ? 'bg-gray-800 text-gray-300 hover:bg-gray-750' : 'bg-gray-50 text-gray-700 hover:bg-gray-100' + }`} + > + <span>Round {round.round} ({round.responses.length} responses)</span> + <div className="flex items-center gap-2"> + {round.judgeDecision && ( + <span className={`px-1.5 py-0.5 rounded text-[10px] ${ + round.judgeDecision.continue + ? isDark ? 'bg-green-900/50 text-green-300' : 'bg-green-100 text-green-700' + : isDark ? 'bg-red-900/50 text-red-300' : 'bg-red-100 text-red-700' + }`}> + {round.judgeDecision.continue ? 'Continue' : 'Stop'} + </span> + )} + {round.converged !== undefined && ( + <span className={`px-1.5 py-0.5 rounded text-[10px] ${ + round.converged + ? isDark ? 'bg-green-900/50 text-green-300' : 'bg-green-100 text-green-700' + : isDark ? 'bg-yellow-900/50 text-yellow-300' : 'bg-yellow-100 text-yellow-700' + }`}> + {round.converged ? 'Converged' : 'Divergent'} + </span> + )} + <ChevronDown size={12} className={`transition-transform ${isExpanded ? 'rotate-180' : ''}`} /> + </div> + </button> + {isExpanded && ( + <div className={`p-2 space-y-2 ${isDark ? 'bg-gray-900' : 'bg-white'}`}> + {round.responses.map((resp, ri) => ( + <div key={ri} className={`rounded border p-2 ${isDark ? 'border-gray-700' : 'border-gray-200'}`}> + <div className={`text-xs font-semibold mb-1 ${isDark ? 'text-cyan-400' : 'text-cyan-600'}`}>{resp.model}</div> + <div className={`text-xs prose prose-sm max-w-none ${isDark ? 'prose-invert text-gray-300' : 'text-gray-700'}`}> + {rawTextMode ? ( + <pre className="whitespace-pre-wrap break-words">{resp.response}</pre> + ) : ( + <ReactMarkdown remarkPlugins={[remarkGfm, remarkMath]} rehypePlugins={[rehypeKatex]}> + {preprocessLaTeX(resp.response)} + </ReactMarkdown> + )} + </div> + </div> + ))} + {round.judgeDecision && ( + <div className={`rounded border p-2 text-xs ${isDark ? 'border-amber-800/50 bg-amber-900/20 text-amber-200' : 'border-amber-200 bg-amber-50 text-amber-800'}`}> + <div className="font-semibold mb-1">Judge Decision</div> + <div className="whitespace-pre-wrap">{round.judgeDecision.reasoning}</div> + </div> + )} + </div> + )} + </div> + ); + })} + {selectedNode.data.status === 'loading' && debateStage && ( + <div className={`flex items-center gap-2 p-2 text-xs ${isDark ? 'text-cyan-400' : 'text-cyan-600'}`}> + <Loader2 className="animate-spin" size={12} /> {debateStage} + </div> + )} + </div> + )} + + {/* Per-Model View tab */} + {debateTab === 'per-model' && ( + <div> + {(() => { + const allModels = new Set<string>(); + selectedNode.data.debateData!.rounds.forEach(r => r.responses.forEach(resp => allModels.add(resp.model))); + const modelList = Array.from(allModels); + const selectedModel = debatePerModelSelected || modelList[0] || ''; + return ( + <> + <div className={`flex gap-0.5 mb-2 flex-wrap`}> + {modelList.map(model => ( + <button + key={model} + onClick={() => setDebatePerModelSelected(model)} + className={`px-2 py-0.5 text-xs rounded transition-colors ${ + (debatePerModelSelected || modelList[0]) === model + ? isDark ? 'bg-cyan-900/50 text-cyan-300' : 'bg-cyan-100 text-cyan-700' + : isDark ? 'bg-gray-800 text-gray-400 hover:bg-gray-700' : 'bg-gray-100 text-gray-500 hover:bg-gray-200' + }`} + > + {model} + </button> + ))} + </div> + <div className="space-y-2 max-h-[400px] overflow-y-auto"> + {selectedNode.data.debateData!.rounds.map(round => { + const resp = round.responses.find(r => r.model === selectedModel); + if (!resp) return null; + return ( + <div key={round.round} className={`rounded border p-2 ${isDark ? 'border-gray-700' : 'border-gray-200'}`}> + <div className={`text-xs font-semibold mb-1 ${isDark ? 'text-gray-400' : 'text-gray-500'}`}>Round {round.round}</div> + <div className={`text-xs prose prose-sm max-w-none ${isDark ? 'prose-invert text-gray-300' : 'text-gray-700'}`}> + {rawTextMode ? ( + <pre className="whitespace-pre-wrap break-words">{resp.response}</pre> + ) : ( + <ReactMarkdown remarkPlugins={[remarkGfm, remarkMath]} rehypePlugins={[rehypeKatex]}> + {preprocessLaTeX(resp.response)} + </ReactMarkdown> + )} + </div> + </div> + ); + })} + </div> + </> + ); + })()} + </div> + )} + </div> ) : isEditing ? ( <div className="space-y-2"> <textarea diff --git a/frontend/src/store/flowStore.ts b/frontend/src/store/flowStore.ts index f31720e..c1c8b99 100644 --- a/frontend/src/store/flowStore.ts +++ b/frontend/src/store/flowStore.ts @@ -91,6 +91,23 @@ export interface CouncilData { stage3: { model: string; response: string } | null; } +export interface DebateRound { + round: number; + responses: Array<{ model: string; response: string }>; + judgeDecision?: { continue: boolean; reasoning: string }; + converged?: boolean; +} + +export interface DebateData { + rounds: DebateRound[]; + finalVerdict: { model: string; response: string } | null; + config: { + judgeMode: 'external_judge' | 'self_convergence' | 'display_only'; + format: 'free_discussion' | 'structured_opposition' | 'iterative_improvement' | 'custom'; + maxRounds: number; + }; +} + export interface NodeData { label: string; model: string; @@ -110,6 +127,16 @@ export interface NodeData { chairmanModel?: CouncilMemberConfig; councilData?: CouncilData; + // Debate mode + debateMode?: boolean; + debateModels?: CouncilMemberConfig[]; + judgeModel?: CouncilMemberConfig; + debateJudgeMode?: 'external_judge' | 'self_convergence' | 'display_only'; + debateFormat?: 'free_discussion' | 'structured_opposition' | 'iterative_improvement' | 'custom'; + debateCustomPrompt?: string; + debateMaxRounds?: number; + debateData?: DebateData; + // Traces logic traces: Trace[]; // INCOMING Traces outgoingTraces: Trace[]; // ALL Outgoing (inherited + self + forks + merged) |