💬AI Assistant

Use the AI assistant for quick help, code queries, and document search

Your intelligent assistant for platform help and knowledge queries

---

📖 Table of Contents

1. Overview

2. Quick Start

3. Chat Interface

4. Message Types

5. Code Snippets

6. Document References

7. Common Tasks

8. Tips & Best Practices

9. Troubleshooting

---

Overview

What is the AI Assistant?

The AI Assistant is your built-in chatbot that can answer questions about:

• Platform features: How to use Automatos AI

• Your documents: Search and query your knowledge base

• Your code: Ask about code in indexed repositories

• Troubleshooting: Get help with errors and issues

Access: Click Assistant in sidebar, or click chat icon (💬) anywhere

AI Assistant Overview

What Can You Ask?

Platform Help:

• "How do I create an agent?"

• "What's the difference between workflows and agents?"

• "How do I upload documents?"

Document Queries:

• "What does our security policy say about passwords?"

• "Summarize the Q4 report"

• "Find deployment procedures"

Code Queries (if CodeGraph enabled):

• "How is authentication implemented?"

• "Where is the payment processing code?"

• "Show me examples of error handling"

Troubleshooting:

• "Why did my workflow fail?"

• "Agent not executing tasks - help!"

• "Document upload stuck - what to do?"

---

Quick Start

Ask Your First Question (1 Minute)

Goal: Get help from AI Assistant

Steps:

1. Click Assistant in sidebar

2. Type question: "How do I create an agent?"

3. Press Enter

4. Get instant response with links

5. Follow links for detailed guides

⏱️ Time: 30 seconds 🎯 Result: Answer with documentation links

---

Chat Interface

Layout

The chat interface has 3 main sections:

Chat Interface Layout

1. Welcome Message (on first load)

💡 Tooltip: "Introduces the assistant and suggests starter questions"

Displays:

• Welcome text

• Suggested questions (click to ask)

• Feature highlights

• Quick tips

Suggested Questions:

• "How do I create my first agent?"

• "What are workflows?"

• "How do I upload documents?"

• "Show me example workflows"

Welcome Message

2. Message History

💡 Tooltip: "Conversation history. Scroll to see previous messages."

Shows:

• All messages in conversation

• Your messages (right side, blue)

• Assistant messages (left side, gray)

• Timestamps

• Code snippets (formatted)

• Document references (with links)

Message Features:

• Auto-scroll to latest

• Copy message text

• Regenerate response (if unsatisfied)

• Share conversation

Message History

3. Input Area

Text Input 💡 Tooltip: "Type your question or message. Press Enter to send, Shift+Enter for new line."

• Multi-line support

• Auto-resize as you type

• Character count (if long)

Send Button 💡 Tooltip: "Send message to assistant. Also: press Enter"

• Disabled while assistant is typing

• Shows spinner during response

• Keyboard shortcut: Enter

Attach Button (optional) 💡 Tooltip: "Attach code snippet or document reference for context"

• Paste code

• Link to document

• Add screenshot (future)

Input Area

---

Message Types

User Messages

Your messages appear:

• Right-aligned

• Blue background

• Your avatar/initials

• Timestamp

Message Actions (hover):

• Copy text

• Edit message (resend)

• Delete message

User Message

Assistant Messages

Assistant messages appear:

• Left-aligned

• Gray background

• AI icon

• Timestamp

Rich Content Support:

Plain Text

Standard text responses with:

• Markdown formatting

• Bold, italic, code

• Lists and bullets

• Paragraphs

Assistant Text Message

Code Blocks

Formatted code with syntax highlighting:

# Example code snippet
def create_agent(name, agent_type):
    agent = Agent(name=name, type=agent_type)
    return agent

Code Block Features:

• Syntax highlighting (via Prism.js)

• Language badge (Python, JavaScript, etc.)

• Copy button (one-click copy)

• Line numbers (optional)

Code Block

Links

Clickable links to:

• Documentation pages

• Agent details

• Workflow results

• Document previews

Link Preview (hover):

• Shows link destination

• Preview tooltip

• Safe link indicator

---

Code Snippets

Code References Panel

When assistant includes code, a side panel may appear:

💡 Tooltip: "Extracted code snippets for easy reference and copying"

Code Snippets Panel

Features:

Code Snippet Cards:

• Title: Description of what code does

• Language: Programming language

• Source: Where code is from (if from CodeGraph)

• Code: Formatted, highlighted code

• Copy Button: Copy to clipboard

Example Snippet Card:

Title: Create Agent Function
Language: Python
Source: agents/agent_factory.py:145

def create_specialized_agent(
    name: str,
    agent_type: str,
    model_config: dict
) -> Agent:
    """Create agent with configuration"""
    ...

Actions:

• 📋 Copy: Copy code to clipboard

• 📄 View Full File: Open source file (if CodeGraph)

• 🤖 Explain: Ask assistant to explain this code

• ✏️ Modify: Get help modifying code

Code Execution (Future Feature)

🔧 Advanced (coming soon)

• Run code snippets in sandbox

• See execution results

• Test before using

• Save to library

---

Document References

Document Panel

When assistant references documents, a panel shows sources:

💡 Tooltip: "Documents referenced in response. Click to view full document."

Document References

Reference Cards:

Each document shows:

• 📄 Document Name

• Page/Section: Where content is from

• Relevance: Similarity score

• Preview: Relevant excerpt

Example Reference:

📄 Security Best Practices.pdf
Page: 15
Relevance: 94%
Preview: "SQL injection prevention requires parameterized 
queries in all database interactions..."

Actions:

• 👁️ View: Open document viewer

• 📋 Copy: Copy excerpt

• 🔍 Search More: Find related content in document

• ⬇️ Download: Download full document

Context Sources

Shows where assistant got information:

Source Types:

• 📚 Knowledge Base: Your uploaded documents

• 💻 CodeGraph: Your indexed code

• 📖 Platform Docs: Built-in documentation

• 🧠 Agent Memory: Learned from past interactions

Transparency:

• All sources cited

• Click to verify

• See similarity scores

• Trust and verify responses

---

Common Tasks

Task 1: Getting Platform Help

Scenario: You don't know how to do something

Steps:

1. Open AI Assistant

2. Ask naturally: "How do I configure credentials for Slack?"

3. Get response with:

   • Step-by-step instructions

   • Links to Settings page

   • Relevant documentation links

4. Follow instructions

5. Ask follow-up if needed: "What permissions do I need?"

⏱️ Time: 1-2 minutes 🎯 Result: Clear guidance

Task 2: Searching Your Documents

Scenario: Find information in your knowledge base

Steps:

1. Open AI Assistant

2. Ask: "What does our API documentation say about rate limiting?"

3. Assistant searches your documents

4. Returns answer with document references

5. Click reference to open full document

6. Get exact information needed

⏱️ Time: 30 seconds 🎯 Result: Found information with source

Task 3: Understanding Your Code

Scenario: Need to understand code in your repository

Prerequisites: CodeGraph project indexed

Steps:

1. Open AI Assistant

2. Ask: "How does user authentication work in our app?"

3. Assistant searches CodeGraph

4. Returns explanation with code references

5. Click code reference to see full file

6. Ask follow-up: "Show me the password hashing function"

⏱️ Time: 1-2 minutes 🎯 Result: Code understanding with examples

Task 4: Troubleshooting Errors

Scenario: Workflow failed with error

Steps:

1. Copy error message from workflow

2. Open AI Assistant

3. Paste error: "Error: Agent timeout after 5 minutes"

4. Get explanation and solutions:

   • What this error means

   • Common causes

   • How to fix

5. Apply suggested solution

6. Ask if unclear: "How do I increase the timeout?"

⏱️ Time: 2-3 minutes 🎯 Result: Error resolved

---

Advanced Features

Conversation Context

🔧 Advanced

The assistant maintains conversation context:

Context Awareness:

• Remembers previous questions in session

• Understands follow-up questions

• References earlier code snippets

• Builds on previous answers

Example:

You: "How do I create an agent?"
Assistant: [Explains agent creation...]

You: "What about adding skills to it?"
Assistant: "To add skills to the agent we just discussed, 
you can..." [Continues context]

Managing Context:

• Clear Context: Start fresh conversation

• Export Conversation: Save chat history

• Bookmark Response: Save specific answer

Multi-Turn Conversations

Complex questions broken into turns:

You: "I need to set up automated code reviews"
Assistant: "I'll help you set up code reviews. What programming 
language is your code?"

You: "Python"
Assistant: "Great! For Python code reviews, I recommend:
1. Create a Code Architect agent with Python skills
2. Add GitHub tool for PR access
3. Create workflow with security checks
Would you like me to walk through each step?"

You: "Yes, start with creating the agent"
Assistant: [Detailed agent creation steps...]

Benefits:

• Break complex tasks into steps

• Interactive guidance

• Contextual follow-ups

Session Management

Sessions persist:

• Conversation saved across page navigation

• Return to continue conversation

• Sessions expire after 24 hours inactivity

Starting New Session:

• Click "New Conversation" button

• Clears current context

• Fresh start

---

Tips & Best Practices

Asking Good Questions

💡 Best Practices:

Good questions:

• ✅ "How do I add Slack integration to an agent?"

• ✅ "What's causing error: 'Agent timeout exceeded'?"

• ✅ "Show me the authentication code in auth_service.py"

Poor questions:

• ❌ "help"

• ❌ "not working"

• ❌ "code"

Question Tips:

1. Be specific about what you need

2. Include error messages (copy-paste)

3. Mention what you've already tried

4. Provide context (which page, which agent, etc.)

Using Code Queries

For CodeGraph queries:

1. Mention file names when known

2. Use technical terms from your codebase

3. Ask about specific functions/classes

4. Follow up for details

Example Flow:

You: "Where is the payment processing code?"
Assistant: [Shows PaymentService in services/payment.py]

You: "How does validate_payment work?"
Assistant: [Shows function code and explanation]

You: "Are there any security issues in that function?"
Assistant: [Analyzes for vulnerabilities]

Getting Document Information

For document queries:

1. Mention document name if known

2. Ask specific questions

3. Request summaries for long docs

4. Ask for specific sections

Example:

You: "Summarize the Q4 financial report"
Assistant: [Provides summary with key metrics]

You: "What was the revenue growth?"
Assistant: "According to the Q4 report (page 3), 
revenue grew 23% year-over-year..."

---

Troubleshooting

Assistant Not Responding

Symptom: Message sent but no response

Solutions:

1. Check connection:

   • Look for connection status indicator

   • Green = connected, Red = disconnected

   • Refresh page if disconnected

2. Wait longer:

   • Complex queries take 5-10 seconds

   • Document searches take 3-5 seconds

   • Code queries take 2-4 seconds

3. Check message length:

   • Very long messages may timeout

   • Break into smaller questions

4. Retry:

   • Click retry button (if appears)

   • Or rephrase and send again

Incorrect or Unhelpful Responses

Symptom: Assistant gives wrong answer

Solutions:

1. Provide more context:

   • Add details about what you're trying to do

   • Mention specific page or feature

   • Include error messages

2. Rephrase question:

   • Try different wording

   • Break into multiple questions

   • Be more specific

3. Verify sources:

   • Check document references

   • Verify code snippets

   • Cross-check with documentation

4. Give feedback:

   • Thumbs up/down on responses

   • Helps improve assistant

No Code References

Symptom: Asked about code but no code shown

Solutions:

1. Check CodeGraph:

   • Is your code indexed?

   • Go to Knowledge → CodeGraph

   • Verify project exists

2. Be specific:

   • Mention file names

   • Include function/class names

   • Ask about specific files

3. Check project selection:

   • Some setups have multiple projects

   • Specify which project

   • "In the backend project, where is..."

No Document References

Symptom: Asked about documents but assistant doesn't find them

Solutions:

1. Check documents processed:

   • Knowledge → Library

   • Verify status is "Processed" (green)

2. Try keyword search:

   • Use exact terms from document

   • Include document name if known

3. Upload document:

   • May not be in knowledge base

   • Upload and wait for processing

   • Try query again

---

Related Guides

• Knowledge Base Guide: Upload documents for chatbot queries

• CodeGraph: Index code for code queries

• Agents Guide: Learn about agents (chatbot can explain)

---

Keyboard Shortcuts

• Ctrl/Cmd + K: Focus chat input

• Enter: Send message

• Shift + Enter: New line in message

• Esc: Close chat (if modal)

• ↑ Arrow: Edit last message

• Ctrl/Cmd + L: Clear conversation

---

FAQ

Is the chatbot connected to my knowledge base?

Yes! The assistant can search:

• Your uploaded documents

• Your indexed code (CodeGraph)

• Platform documentation

• Agent memories

How it works:

1. You ask a question

2. Assistant searches knowledge base

3. Retrieves relevant content

4. Generates answer with citations

What language models power the assistant?

Default: GPT-4 Turbo for best quality

Can be configured in Settings → System Settings → Chatbot Configuration

Options:

• GPT-4 Turbo: Best quality, moderate speed

• GPT-4: High quality, slower

• GPT-3.5 Turbo: Fast, lower cost

Does it remember conversations?

Within session: Yes

• Conversation context maintained

• Follow-up questions work

• References previous messages

Across sessions: No (currently)

• Future feature: conversation history

• Workaround: Export important conversations

Can I disable the assistant?

Yes, in Settings → System Settings → Features:

• Disable chatbot toggle

• Removes from navigation

• Clears chat data (optional)

How accurate are the responses?

Generally high accuracy when:

• ✅ Asking about platform features (>95%)

• ✅ Querying your documents (>90%)

• ✅ CodeGraph queries (>85%)

Less accurate when:

• ⚠️ Complex multi-step questions

• ⚠️ Very recent features

• ⚠️ Ambiguous questions

Always:

• Check sources

• Verify code snippets

• Test suggestions before applying

---

Next: Tools Guide →

Learn to integrate 400+ tools with your agents