Chat Assistant

The Chat Assistant provides conversational AI interfaces with persistent memory, thread management, and tool integration. Build intelligent chatbots for customer support, internal Q&A, document analysis, and knowledge exploration.

Chat Features

M3 Forge’s Chat Assistant combines multiple AI capabilities:

Thread-Based Conversations

• Persistent history — Conversations saved across sessions
• Thread management — Create, archive, delete threads
• Context continuity — Assistant remembers entire conversation
• Thread switching — Resume any past conversation

Memory Integration (mem0)

• Long-term memory — Remember user preferences and facts across threads
• Contextual awareness — Recall relevant information automatically
• Personalization — Adapt responses based on user history
• Memory toggle — Enable/disable per thread

Assistant Selection

• Specialized assistants — Choose assistant for specific tasks
• Pre-configured skills — Assistants come with relevant tools
• Custom personalities — Different tone and behavior per assistant
• Quick switching — Change assistant mid-conversation

Image Support

• Image attachments — Upload images to messages
• Visual Q&A — Ask questions about images
• Multi-modal understanding — Combine text and images
• Drag-and-drop — Easy image upload

Skill/Tool Integration

• Command palette (Cmd+K) — Invoke tools during conversation
• Available skills — Access processor, workflow, and knowledge capabilities
• Tool results inline — See tool outputs in conversation
• Skill management — Enable/disable skills per thread

Real-Time Streaming

• Token-by-token — See response generate in real-time
• Interruption — Stop generation if needed
• Progress indication — Show when assistant is thinking

Using Chat Assistant

Access Chat from Apps → Chat in main navigation.

Starting a Conversation

Thread Management

Create and organize conversation threads:

Creating Threads:

• Click New Chat to start fresh thread
• Existing messages saved to current thread
• New thread starts with clean context

Switching Threads:

• Click sidebar to view thread list
• Select thread to resume conversation
• Recent threads at top

Archiving Threads:

• Right-click thread → Archive
• Archived threads hidden from main list
• Access via “Show Archived” toggle

Deleting Threads:

• Right-click thread → Delete
• Confirm deletion
• Permanently removes thread and messages

Memory (mem0)

Memory enables personalization across conversations:

What Memory Stores

• User preferences — Settings, favorites, dislikes
• Facts about user — Name, role, organization details
• Context — Projects, documents, topics of interest
• Conversation patterns — Frequently asked questions

Enabling Memory

Memory is enabled by default. Toggle per thread:

1. Click memory indicator in header (brain icon)
2. Toggle “Enable Memory” switch
3. Changes apply to current thread

When to Enable Memory:

• Personal assistant use
• Long-term project work
• Repeated interactions
• Personalized recommendations

When to Disable Memory:

• Sensitive/private queries
• One-off questions
• Shared accounts
• Testing/experimentation

Memory Privacy

Memory data is:

• User-scoped — Only accessible to you
• Thread-scoped — Isolated per conversation
• Deletable — Clear memory anytime
• Encrypted — Stored securely

Clear memory via Settings → Clear Memory.

Image Support

Attach images for visual questions:

Uploading Images

Methods:

• Click image upload button
• Drag-and-drop onto chat
• Paste from clipboard (Ctrl/Cmd+V)

Supported Formats:

• PNG, JPEG, WEBP
• Maximum size: 20MB
• Multiple images per message

Image Preview

Before sending:

• Thumbnails show attached images
• Click to view full size
• Remove with X button
• Reorder by dragging

Visual Q&A

Ask questions about images:

• “What’s in this image?”
• “Extract text from this screenshot”
• “Identify objects in this photo”
• “What’s wrong with this diagram?”
• “Transcribe this handwritten note”

Assistant analyzes image and responds with insights.

Skills and Tools

Extend chat capabilities with skills:

Available Skills

Skills depend on selected assistant. Common skills:

Invoking Skills

Via Command Palette (Cmd+K):

1. Press Cmd+K or Ctrl+K
2. Palette opens with skill list
3. Type to filter skills
4. Select skill
5. Skill tag appears in message input
6. Type additional context
7. Send message

Via Slash Commands:

Type / in message input to see skill list. Select skill from autocomplete.

Via Natural Language:

Ask assistant naturally:

• “Search the knowledge base for invoices from ACME Corp”
• “Run the contract analysis workflow on this document”
• “Extract the invoice number and total”

Assistant infers skill to use.

Skill Results

Skill outputs appear in conversation:

• Structured data — Tables or formatted display
• Documents — Links to download or view
• Status updates — Workflow execution progress
• Errors — If skill fails, error message shown

Managing Skills

Toggle skills per thread:

1. Click skills indicator in header
2. View active skills
3. Enable/disable specific skills
4. Changes apply to current thread

Disabling skills prevents assistant from using them.

Suggested Prompts

Assistants provide suggested prompts to guide conversation:

When Shown

Suggested prompts appear:

• At conversation start (empty thread)
• After assistant selection
• When assistant is idle

Using Prompts

Click suggested prompt to:

• Fill message input with prompt
• Optionally edit prompt
• Send to assistant

Prompts showcase assistant capabilities and common use cases.

Custom Prompts

Administrators can configure custom prompts per assistant:

1. Navigate to Admin → Assistants
2. Edit assistant
3. Add suggested prompts
4. Save changes

Prompts help users discover features.

Assistant Configuration

Customize assistant behavior:

Creating Assistants

System Prompts

System prompts define assistant personality and behavior:

Example:

You are a helpful contract analysis assistant. Your role is to:
- Review contract documents for key terms and risks
- Explain legal language in plain English
- Identify missing or unusual clauses
- Provide recommendations for negotiation

Always:
- Be professional and precise
- Cite specific contract sections
- Explain your reasoning
- Ask clarifying questions when needed

Never:
- Provide legal advice (recommend consulting a lawyer)
- Make decisions on behalf of the user
- Assume information not in the contract

Craft prompts carefully to ensure desired behavior.

Knowledge Base Integration

Connect assistants to knowledge base for RAG:

Configuring Knowledge Access

1. Navigate to assistant settings
2. Select Knowledge Base tab
3. Choose RAG indices to include
4. Configure search settings:
   • Number of results to retrieve
   • Similarity threshold
   • Metadata filters

How RAG Works

When user asks question:

1. Assistant searches knowledge base
2. Retrieves relevant document chunks
3. Uses chunks as context for response
4. Cites sources in answer

Provides accurate, grounded responses.

Citation Display

Responses with knowledge base sources show:

• Inline citations (e.g., [1], [2])
• Reference list at end of response
• Links to source documents
• Confidence scores

Users can verify information.

Advanced Features

Multi-Assistant Workflows

Switch assistants mid-conversation:

1. Use general assistant for initial query
2. Switch to specialist for deep dive
3. Return to general for follow-up

Context carries over to new assistant.

Conversation Export

Export thread for archival or sharing:

1. Click thread menu → Export
2. Choose format (Markdown, PDF, HTML)
3. Download file

Includes all messages and metadata.

Conversation Sharing

Share thread with team:

1. Click thread menu → Share
2. Generate shareable link
3. Set permissions (view, comment, edit)
4. Send link to collaborators

Useful for troubleshooting or collaboration.

Voice Input

Speak messages instead of typing:

1. Click microphone button
2. Speak message
3. Speech-to-text transcription appears
4. Edit if needed, then send

Requires browser microphone permissions.

Mobile Experience

Chat works seamlessly on mobile:

• Responsive layout — Optimized for phone screens
• Touch gestures — Swipe to close, long-press for options
• Offline support — Threads cached for offline viewing
• Push notifications — Alert when assistant responds (if installed as PWA)

Install M3 Forge as PWA for best mobile experience.

Troubleshooting

No LLM Configured

Symptom: Warning message about missing LLM configuration.

Solution:

1. Navigate to Admin → LLM Connections
2. Add connection (OpenAI, Anthropic, etc.)
3. Test connection
4. Refresh chat

Assistant Not Responding

Symptoms: Messages send but no response.

Solutions:

• Check LLM connection status
• Verify API key is valid and has credits
• Review rate limits
• Check for errors in browser console

Memory Not Working

Symptoms: Assistant doesn’t remember preferences.

Solutions:

• Verify memory is enabled for thread
• Check mem0 service is running
• Review memory storage permissions
• Clear and rebuild memory

Slow Responses

Symptoms: Responses take long time to generate.

Solutions:

• Check LLM API status
• Reduce max tokens
• Simplify system prompt
• Use faster model (GPT-3.5 vs GPT-4)

Best Practices

Effective Prompting

• Be specific — Provide context and details
• Ask one thing — Break complex queries into steps
• Iterate — Refine based on responses
• Provide examples — Show desired format

Context Management

• Use threads — Separate topics into different threads
• Prune context — Start new thread when conversation gets long
• Reference earlier — Refer to previous messages by content

Privacy

• Don’t share sensitive data — Avoid PII, credentials, secrets
• Disable memory — For sensitive queries
• Delete threads — Remove threads with sensitive info
• Review sharing — Check who has access to shared threads

Next Steps

• Build form-based apps with App Builder
• Create Custom Apps with advanced features
• Integrate external systems via Webhooks
• Connect chat to Knowledge Base for RAG