Spaces:
Sleeping
🔧 Fix: Don't Reset Chat When Generating New Agent
Problem Identified
When a user was in the middle of a conversation with an agent in the builder panel and then generated a new agent, the chat would reset, interrupting the ongoing conversation.
User Experience (Before - Broken):
User: "What antibiotics for pneumonia?"
Agent: "I recommend ceftriaxone and azithromycin..."
User: "What about allergies?"
[User generates a new agent]
❌ Chat resets to: "📋 Agent generated successfully! Please select an agent to begin testing."
❌ Previous conversation lost
❌ Chat controls disabled
Root Cause
The generate_button.click() event had a chain of .then() steps, and one of them was calling show_initial_instruction_state() which:
- Replaced the chatbot content with instruction message
- Disabled the chat input box
- Disabled send and reset buttons
- Cleared invocation logs and active children
# OLD (Broken):
generate_button.click(...)
.then(handle_generate, ...) # Generate agent
.then(show_initial_instruction_state, ...) # ❌ RESET CHAT - interrupts conversation
.then(auto_load_fields, ...) # Load new agent fields
.then(refresh_dropdowns, ...) # Update agent lists
This was originally designed when agents were auto-loaded into chat after generation, but the workflow changed to require users to manually select an agent to chat with.
Solution Applied
Removed the .then() step that resets the chat, while keeping all other functionality:
# NEW (Fixed):
generate_button.click(...)
.then(handle_generate, ...) # Generate agent
# ✅ REMOVED: .then(show_initial_instruction_state, ...)
.then(auto_load_fields, ...) # Load new agent fields into form
.then(refresh_dropdowns, ...) # Update agent lists and dropdowns
.then(show_success_message, ...) # Show "✅ Agent generated successfully!"
Code Change:
File: app.py lines 2152-2167
Removed:
).then( # 2) show initial instruction instead of preload demo chat
lambda: show_initial_instruction_state(),
inputs=[], outputs=[builder_chatbot, chat_input, builder_send_button, reset_button, invocation_log, active_children]
How It Works Now
User Experience (After - Fixed):
User: "What antibiotics for pneumonia?"
Agent A: "I recommend ceftriaxone and azithromycin..."
User: "What about allergies?"
[User generates Agent B while still chatting with Agent A]
✅ Chat continues uninterrupted with Agent A
✅ Agent B is created in the background
✅ Agent B appears in dropdown
✅ Success message: "✅ Agent generated successfully!"
✅ User can continue conversation with Agent A
✅ User can switch to Agent B when ready
What Still Works
After generating an agent, these actions still happen automatically:
✅ Agent is saved to user's session
✅ Agent fields auto-load into the builder form (type, name, mission, skills)
✅ Active agents list refreshes showing the new agent
✅ Dropdown updates with the new agent name
✅ Success message displays "✅ Agent generated successfully!"
✅ Chat dropdown refreshes in the deployed chat panel
What Changed
❌ Chat no longer resets after agent generation
❌ Chat controls stay enabled if you were chatting
❌ Conversation history preserved during agent creation
❌ Current agent selection unchanged when generating new agent
Use Cases Enabled
Use Case 1: Build Agents During Consultation
Scenario: User is consulting with "General ID Agent" and realizes they need a specialist
1. User chats with "General ID Agent" about complex case
2. Agent suggests: "You might want to consult a stewardship specialist"
3. User generates "Stewardship Expert" agent WITHOUT leaving current chat
4. User continues with "General ID Agent" while new agent is created
5. User can switch to "Stewardship Expert" when ready
Use Case 2: Iterative Agent Development
Scenario: User is testing an agent and wants to create variations
1. User chats with "ID Agent v1" to test its responses
2. User identifies improvements needed
3. User generates "ID Agent v2" with enhanced skills
4. User can compare by switching between v1 and v2
5. Chat history with v1 is preserved for reference
Use Case 3: Multi-Agent Workflow Setup
Scenario: User is building an orchestrator + subagents
1. User creates orchestrator agent first
2. User tests orchestrator to understand what subagents are needed
3. While viewing orchestrator's response, user generates subagent 1
4. User generates subagent 2, 3, 4 without interrupting workflow
5. User can test orchestrator again when all subagents are ready
Testing
Quick Test:
Start a conversation:
- Select an agent from dropdown
- Click "Chat with Selected Agent"
- Send a message: "What antibiotics for UTI?"
- Get a response from the agent
Generate a new agent:
- Fill in agent builder form (name, type, mission, skills)
- Click "Generate Agent"
- Wait for success message
Verify:
- ✅ Chat history still visible
- ✅ Previous conversation preserved
- ✅ Can continue chatting with current agent
- ✅ New agent appears in dropdown
- ✅ Can switch to new agent when desired
Files Modified
| File | Change | Lines |
|---|---|---|
app.py |
Removed chat reset step from generate_button.click | -3 lines |
PUBMED_SETUP.md |
Added (documentation for PubMed setup) | +218 lines |
Deployment Status
✅ Fixed and Deployed
- Commit:
c10e5be - Message: "Fix: Don't reset chat when generating new agent - Allow agent creation during ongoing conversations"
- Space: https://huggingface.co/spaces/John-jero/IDWeekAgents
- Status: LIVE
Impact
✅ Better UX - No interruption of ongoing conversations
✅ More Flexible - Build agents while consulting
✅ Multi-Agent Workflows - Easier to set up orchestrator + subagents
✅ No Breaking Changes - All other functionality preserved
Related Functionality
The chat can still be reset when needed:
✅ Manual Reset: Click "🔄 Reset Chat" button
✅ Switch Agent: Select different agent and click "Chat with Selected Agent"
✅ New Session: Browser refresh clears in-memory data
Chat interruptions during agent generation are now eliminated! 🎉
Users can seamlessly build agents while actively consulting with other agents.