diff --git a/fern/docs.yml b/fern/docs.yml index b763c7d..d70957c 100644 --- a/fern/docs.yml +++ b/fern/docs.yml @@ -17,6 +17,9 @@ tabs: development: display-name: Development and Integration icon: book + agent-forge: + display-name: Agent Forge + icon: fa-solid fa-wand-magic-sparkles api: display-name: API Reference icon: puzzle @@ -33,6 +36,20 @@ navigation: path: ./docs/pages/welcome.mdx - page: Security path: ./docs/pages/security.mdx + - section: Agent Forge + contents: + - page: Overview + path: ./docs/pages/agent-forge/overview.mdx + - page: Getting Started + path: ./docs/pages/agent-forge/getting-started.mdx + - page: Chatting with Agents + path: ./docs/pages/agent-forge/chatting.mdx + - page: Building Agents + path: ./docs/pages/agent-forge/building-agents.mdx + - page: Knowledge Base + path: ./docs/pages/agent-forge/knowledge-base.mdx + - page: MCP Integration + path: ./docs/pages/agent-forge/mcp.mdx - tab: deployment layout: - section: Getting Started @@ -143,7 +160,24 @@ navigation: contents: - page: Contact Us path: ./docs/pages/support.mdx - + - tab: agent-forge + layout: + - section: Getting Started + contents: + - page: Overview + path: ./docs/pages/agent-forge/overview.mdx + - page: Getting Started + path: ./docs/pages/agent-forge/getting-started.mdx + - section: Using Agent Forge + contents: + - page: Chatting with Agents + path: ./docs/pages/agent-forge/chatting.mdx + - page: Building Agents + path: ./docs/pages/agent-forge/building-agents.mdx + - page: Knowledge Base + path: ./docs/pages/agent-forge/knowledge-base.mdx + - page: MCP Integration + path: ./docs/pages/agent-forge/mcp.mdx - tab: api layout: - api: API Reference diff --git a/fern/docs/pages/agent-forge/building-agents.mdx b/fern/docs/pages/agent-forge/building-agents.mdx new file mode 100644 index 0000000..25234f8 --- /dev/null +++ b/fern/docs/pages/agent-forge/building-agents.mdx @@ -0,0 +1,453 @@ +--- +title: Building Agents +subtitle: Agent Forge enables you to create specialized AI agents tailored to your organization's needs. This comprehensive guide covers all aspects of agent creation; from basic setup to advanced configurations. +description: Complete guide to building agents with custom models, prompts, knowledge bases, and security settings +--- + +## Agent Builder Overview + +Access the Agent Builder from the left sidebar to create and manage your agents. + +![Agent Builder Screen](./screenshots/03-01-Agent-builder-screen.png) + +The Agent Builder provides: +- **Your Agents**: Personal agents you've created +- **Shared Agents**: Organization-wide agents +- **Create New Agent**: Button to start building +- **Search**: Find agents by name or description +- **Agent Management**: Edit, duplicate, or delete agents + +## Creating Your First Agent + +### Starting the Creation Process + +![Create New Agent Button](./screenshots/03-02-01-Creating-agent_Create-new-agent-button.png) + +1. **Click "Create New Agent"** in the top-right corner +2. You'll be directed to the **agent configuration form** +3. The form is organized into **sections** for different aspects of your agent + + +**Tip**: Plan your agent's purpose before starting. Consider what tasks it will perform, what knowledge it needs, and who will use it. + + +## Basic Information + +The first section covers fundamental agent details. + +![Basic Information Section](./screenshots/03-02-02-Creating-agent_Basic-info-sys-prompt-streaming-mode.jpg) + +### Agent Name + +**Purpose**: A clear, descriptive name for your agent + +**Best Practices**: +- Use **descriptive names** that indicate the agent's purpose +- Include **domain or function**: "FinServ Compliance Sentry", "Code Review Assistant" +- Keep it **concise but meaningful**: 2-5 words ideal +- Consider **naming conventions** for your organization + +### Description + +**Purpose**: Explain what your agent does and when to use it + +**What to Include**: +- **Primary function**: What the agent specializes in +- **Use cases**: When to use this agent +- **Capabilities**: What it can help with +- **Limitations**: What it cannot do + +## Model & Instructions + +This section defines your agent's intelligence and behavior. + +### Base Model Selection + +**Purpose**: Choose the underlying large language model + +**Available Models** (vary by deployment): +- **gemma-3-27b-it**: Google’s instruction-tuned open-weight model built for balanced performance across diverse tasks. +- **gpt-oss-120b**: OpenAI’s open-weight model designed for powerful reasoning, agentic tasks, and versatile developer use cases. +- **Hermes-3-Llama-3.1-8B**: General use model that excels at reasoning and multi-turn conversations, with an improved focus on longer context lengths. +- **Qwen2.5-Coder-14B-Instruct**: Designed to enhance code generation, reasoning, and fixing, making it a powerful tool for developers. +- **Qwen2.5-VL-7B-Instruct**: Multimodal vision-language mode that excels in analyzing images, video and text. + +**Choosing the Right Model**: + +| Use Case | Recommended Model | Reason | +|----------|-------------------|---------| +| Quick responses, high volume | Smaller models (8B) | Faster inference, lower cost | +| Complex reasoning | Larger models (27B+) | Better understanding, nuanced responses | +| Code generation | Code-specialized models | Better syntax and logic | + + +**Tip**: Start with a mid-size model and adjust based on performance. Smaller models are often sufficient for focused tasks with good prompts. + + +### System Prompt + +**Purpose**: Define your agent's personality, expertise, and behavior + +**The Most Important Configuration**: Your system prompt determines how your agent thinks and responds. + +#### System Prompt Structure + +A well-crafted system prompt includes: + +**1. Role Definition** +``` +You are a financial services compliance expert specializing in +regulatory requirements for banking and fintech companies. +``` + +**2. Core Responsibilities** +``` +Your primary responsibilities: +- Answer questions about financial regulations (SOX, GDPR, PCI DSS) +- Assess compliance risks in business processes +- Provide guidance on regulatory requirements +- Suggest compliance best practices +``` + +**3. Knowledge Scope** +``` +You have deep expertise in: +- US financial regulations (Dodd-Frank, SOX, BSA/AML) +- European regulations (GDPR, MiFID II, PSD2) +- Payment card industry standards (PCI DSS) +- Data protection and privacy laws +``` + +**4. Behavioral Guidelines** +``` +When responding: +- Always cite specific regulations when applicable +- Highlight potential risks clearly +- Provide actionable recommendations +- Acknowledge when unsure and suggest consulting legal counsel +- Use clear, professional language +``` + +**5. Limitations** +``` +Important limitations: +- You are not a lawyer; responses are informational, not legal advice +- Always recommend consulting qualified legal professionals for final decisions +- You cannot guarantee regulatory compliance +``` + +#### System Prompt Best Practices + +**Be Specific**: +❌ "You are a helpful assistant" +✅ "You are a cybersecurity expert specializing in OWASP Top 10 vulnerabilities and secure coding practices for web applications" + +**Define Expertise Boundaries**: +``` +You specialize in Python and JavaScript security. +For questions about other languages, acknowledge your +limitations and suggest consulting language-specific experts. +``` + +**Set Response Format**: +``` +Structure your responses as: +1. Summary (2-3 sentences) +2. Detailed Analysis +3. Recommendations (numbered list) +4. References (if using knowledge base) +``` + +**Include Safety Guidelines**: +``` +Never provide information that could: +- Bypass security controls +- Violate privacy regulations +- Harm users or systems +- Violate ethical guidelines +``` + +**Use Examples in Prompts**: +``` +When explaining code vulnerabilities, provide: +- Example vulnerable code +- Explanation of the risk +- Corrected secure code +- Prevention strategies +``` + +### Streaming Mode + +**Purpose**: Control how responses are delivered + +**Options**: +- **Streaming**: Responses appear word-by-word in real-time +- **Non-Streaming**: Complete response delivered at once + +**When to Use Streaming**: +- ✅ Interactive conversations +- ✅ Long-form responses +- ✅ Better perceived performance +- ✅ User can see progress + +**When to Use Non-Streaming**: +- ✅ Short, quick responses +- ✅ API integrations +- ✅ Batch processing +- ✅ When complete response needed before action +- ✅ When using MCP tool calls + + +**MCP requires "Non-streaming (with MCP capabilities)" mode**: If you plan to connect your agent to external tools via [MCP Integration](/agent-forge/using-agent-forge/mcp), make sure you choose the right mode. MCP tool calls execute in real-time and stream results back as they arrive. + + +## Generation Settings (Advanced Settings) + +Fine-tune how your agent generates responses. + +![Generation Settings](./screenshots/03-02-02-Creating-agent_Generation-settings.jpg) + +### Temperature + +**Range**: 0.0 to 1.0 + +**Purpose**: Controls randomness and creativity in responses + +**Scale**: +``` +0.0 ←────────────────────→ 1.0 +More Deterministic More Creative +``` + +**Recommended Settings**: + +| Use Case | Temperature | Reasoning | +|----------|-------------|-----------| +| Code generation | 0.1 - 0.3 | Need precise, correct syntax | +| Technical documentation | 0.2 - 0.4 | Accuracy over creativity | +| Customer support | 0.3 - 0.5 | Balanced responses | +| Creative writing | 0.7 - 0.9 | Encourage variety | +| Brainstorming | 0.8 - 1.0 | Maximum diversity | +| Data extraction | 0.0 - 0.2 | Strict adherence to format | + + +**Important**: For factual, compliance, or safety-critical applications, use **lower temperature** (0.0-0.3) to ensure consistent, accurate responses. + + +### Top P (Nucleus Sampling) + +**Range**: 0.0 to 1.0 + +**Purpose**: Controls diversity by limiting token choices to a cumulative probability threshold + +**Scale**: +``` +0.0 ←────────────────────→ 1.0 +More Focused More Diverse +``` + +**How It Works**: +- Model ranks all possible next tokens by probability +- Top P selects from tokens that cumulatively reach the threshold +- **Lower values** = Consider only most likely tokens (focused) +- **Higher values** = Consider broader range of tokens (diverse) + +**Recommended Settings**: +- **0.9-0.95**: Good balance for most applications (default) +- **0.5-0.7**: More focused, consistent responses +- **0.95-1.0**: Maximum diversity for creative tasks + +### Top K + +**Range**: 1 to unlimited (typically 1-100) + +**Purpose**: Limit sampling to the K most likely tokens + +**How It Works**: +- Model considers only the K most probable next tokens +- **Lower K** = More focused, repetitive +- **Higher K** = More variety, potential for surprises + +**Recommended Settings**: +- **1**: Always pick most likely token (deterministic) +- **10-20**: Focused, consistent responses +- **40-50**: Balanced variety +- **80-100**: High diversity + +### Max Output Tokens + +**Range**: Varies by model (typically 100-4000+) + +**Purpose**: Maximum number of tokens the model can generate in a single response + +**Understanding Tokens**: +- Tokens ≠ Words +- **Approximate**: 1 token ≈ 0.75 words (English) +- Example: "Hello, world!" ≈ 3-4 tokens + +**Recommended Settings**: + +| Use Case | Max Tokens | Reasoning | +|----------|------------|-----------| +| Short answers | 100-300 | Quick facts, definitions | +| Standard responses | 500-1000 | Detailed explanations | +| Long-form content | 1500-2500 | Articles, documentation | +| Code generation | 1000-2000 | Complete functions/classes | +| Analysis reports | 2000-4000 | Comprehensive reviews | + +## Resources & Knowledge + +Configure how your agent accesses and uses organizational knowledge. + +![Knowledge Base Settings](./screenshots/03-02-03-Creating-agent_Knowledge-base.png) + +### Allow File Uploads + +**Purpose**: Enable users to upload files during conversations + +**Toggle**: On/Off + +**When Enabled**: +- Users can upload documents while chatting +- Agent can read and analyze file contents +- Files persist throughout the conversation + +### Knowledge Bases + +**Purpose**: Attach organizational knowledge to your agent + +**Dropdown**: Select one or more knowledge bases + +**How It Works**: +1. User asks a question +2. System performs **semantic search** on attached knowledge bases +3. Relevant documents/chunks retrieved +4. Context added to agent's prompt +5. Agent generates response using knowledge + +### Customize Search Settings + +**Toggle**: On/Off + +When enabled, you can customize: + +#### Minimum Relevance Score + +**Range**: Off to Stricter + +**Purpose**: Only include results above this similarity score + +**Settings**: +- **Off**: Return all results, no filtering +- **Low (0.3-0.5)**: Include loosely related documents +- **Medium (0.5-0.7)**: Balance precision and recall +- **High (0.7-0.85)**: Only highly relevant documents +- **Strict (0.85-1.0)**: Only near-exact matches + +#### Maximum Results + +**Range**: 1 to 20 + +**Purpose**: Maximum number of document chunks to include in each search + +**Recommended Settings**: +- **1-3**: Focused, specific queries +- **5-10**: Standard (balanced) +- **10-20**: Complex queries needing broad context + +#### Maximum Context Size + +**Range**: Typically 1,000 to 50,000+ characters + +**Purpose**: Maximum characters from knowledge base results to include in agent context + +**Example**: 18000 characters ≈ 4,500 tokens ≈ 3,000 words + +## Access Control & Security + +Configure who can use your agent and what security features are enabled. + +![Access Control and Security](./screenshots/03-02-04-Creating-agent_Access-control-security.jpg) + +### Share With Organization + +**Toggle**: On/Off + +**Purpose**: Make this agent visible to all users in your organization + +**When to Share**: +- ✅ General-purpose assistants +- ✅ Company-wide knowledge agents +- ✅ Standard operating procedure guides + +### Prompt Injection Protection + +**Toggle**: On/Off + +**Purpose**: Detect and block prompt injection attacks in user messages + +**What It Protects Against**: +- Users trying to override system prompt +- Attempts to extract system instructions +- Malicious prompt manipulation +- Jailbreak attempts + +### PII Detection & Replacement + +**Toggle**: On/Off + +**Purpose**: Automatically detect and replace personally identifiable information in messages + +**What It Detects**: +- Names, email addresses, phone numbers +- Social Security Numbers +- Credit card numbers +- Addresses and IP addresses +- Dates of birth + +## Topic Filter + +Control what subjects your agent can discuss. + +![Topic Filter](./screenshots/03-02-05-Creating-agent_Create-agent-page_topic-filter.png) + +### Purpose + +Topic filters ensure your agent stays focused on its intended purpose and doesn't respond to off-topic or inappropriate queries. + +### Configuration Options + +**Allowed Topics** (Whitelist approach): +- Define specific topics agent should address +- Agent politely declines off-topic requests + +**Blocked Topics** (Blacklist approach): +- Define topics agent should refuse +- Agent declines these even if otherwise relevant + +## Saving and Testing Your Agent + +### Saving Your Agent + +1. **Review all settings** one final time +2. **Click "Create Agent"** or "Save Changes" +3. Agent is now available for use +4. Appears in "Your Agents" section + +### Testing Your Agent + +After creation, thoroughly test your agent: + +**Basic Functionality**: +1. Start a new chat with your agent +2. Ask typical questions from your use case +3. Verify responses are appropriate +4. Check tone and style match expectations + +## Next Steps + +Now that you can build agents: + +1. **[Configure MCP](/agent-forge/using-agent-forge/mcp)** - Connect your agent to external tools and live data sources +2. **[Create Knowledge Bases](/agent-forge/using-agent-forge/knowledge-base)** - Add organizational knowledge to your agents +3. **[Share with Team](/agent-forge/using-agent-forge/chatting-with-agents)** - Enable organization sharing and collaborate \ No newline at end of file diff --git a/fern/docs/pages/agent-forge/chatting.mdx b/fern/docs/pages/agent-forge/chatting.mdx new file mode 100644 index 0000000..3a71988 --- /dev/null +++ b/fern/docs/pages/agent-forge/chatting.mdx @@ -0,0 +1,216 @@ +--- +title: Chatting with Agents +subtitle: Agent Forge provides an intuitive chat interface for interacting with AI agents. This guide covers everything from selecting an agent to managing conversations and uploading files. +description: Complete guide to the chat interface, selecting agents, managing conversations, and uploading files +--- + +## The Chat Interface + +### First Screen: Agent Selection + +When you log in, you'll see the main Agent Forge interface with available agents that you have built or shared within your organizations. + +![Agent Selection Screen](./screenshots/01_00_chatting_with_agents_first_screen_select_agent.png) + +## Starting a Conversation + +### Selecting an Agent + +1. **Click the agent dropdown** in the main chat area +2. **Browse available agents** - both your personal agents and shared agents +3. **Click on an agent** to select it +4. The chat interface will update to show the selected agent's name + +### Understanding Agent Cards + +In the Agent Builder view, you'll see agents displayed as cards: + +![Agent Builder Screen](./screenshots/03-01-Agent-builder-screen.png) + +Each agent card shows: +- **Agent Name**: Descriptive title +- **Model**: The underlying LLM (e.g., `gpt-oss-120b`, `gemma-2-27b-it`) +- **Description**: Brief explanation of the agent's purpose +- **Capabilities**: Tags indicating special features (e.g., `shared`, `file-uploads`) +- **Action Buttons**: Options to chat, edit, or manage the agent that you have access to + +### Beginning Your Chat + +Once an agent is selected: + +![Begin Conversation](./screenshots/01_01_begin_conversation.png) + +1. **Review the agent information** displayed at the top +2. **Type your message** in the text box at the bottom +3. **Press Enter** or **click the send button** to submit +4. **Wait for the response** - responses may stream in real-time if streaming mode is enabled + +## Managing Conversations + +### Chat Management Features + +![Chat Management](./screenshots/01_02_chat_management.png) + +The left sidebar provides conversation management tools: + +**Navigation Options**: +- **Starred Conversations**: Quick access to important chats +- **Search Functionality**: Find specific conversations by content or date +- **Chat History**: Browse all previous conversations +- **New Chat**: Start fresh conversation threads + +**Organizing Chats**: +- **Star important conversations**: Click the star icon to mark key discussions +- **Search by keyword**: Use the search bar to find specific topics +- **Delete old chats**: Remove conversations you no longer need + +### Conversation Context + +Each chat maintains context throughout the conversation: +- **Previous messages** are remembered within the session +- **File attachments** remain accessible during the conversation +- **Agent configuration** stays consistent throughout the chat + + +**Best Practice**: Start a new chat when switching topics to ensure clean context and better responses. + + +## Uploading Files + +Agent Forge supports file uploads to provide context and documentation to agents. + +### Upload Methods + +#### Method 1: Click to Upload + +![File Upload Button](./screenshots/02_01_file-upload.png) + +1. **Click the "Upload Files" button** at the bottom of the chat interface +2. **Browse your files** in the file picker dialog +3. **Select one or multiple files** +4. **Click "Open"** to attach files +5. Files will appear in the chat as attachments + +#### Method 2: Drag and Drop + +![Drag and Drop Upload](./screenshots/02_02_drag-and-drop-file-upload.png) + +1. **Select files** from your file manager +2. **Drag files** over the chat interface +3. **Drop files** in the designated area +4. Files are automatically uploaded and attached + +### Supported File Types + +Agent Forge supports various document formats: + +- `.txt` - Plain text files +- `.md` - Markdown files +- `.csv` - Comma-separated values +- `.json` - JSON data files +- `.pdf` - PDF documents +- `.docx` or `.doc` - Microsoft Word documents + + +### Working with Uploaded Files + +Once files are uploaded: + +**When Asking Questions About Files**: +- Make sure to mention the filename (does not have to be exact) that you have uploaded +``` +"Summarize the key points in "KPI-2025.docx"." +"What are the main key points in the PDF file OWASP presentation? (filename: 04-OWASP_presentation.pdf)" +``` + +**File Context**: +- Agents can read and analyze uploaded files +- Multiple files can be referenced in a single conversation +- File content is available throughout the chat session + +**File Management**: +- **View uploaded files**: See list of attachments in the chat +- **Remove files**: Click the X icon on unwanted attachments +- **Updating files**: Remove the old file and upload a new version + +## Best Practices + +### Getting Quality Responses + +**Be Specific**: +❌ "Tell me about security" +✅ "What are the top 3 security vulnerabilities in REST APIs and how can I prevent them?" + +**Provide Context**: +❌ "How do I optimize this?" +✅ "I'm running a Python Flask API serving 10,000 requests/day. How can I optimize database query performance?" + +**Use Follow-ups**: +- Build on previous responses +- Ask for clarification +- Request specific examples or formats + +### File Upload Tips + +**Prepare Files**: +- Use **clear file names** that indicate content +- **Remove sensitive information** before uploading +- **Compress large files** when possible + +**Context in Prompts**: +``` +"I've uploaded our Q4 financial report (Q4-Fin-Report.csv). Can you identify the top 3 expense categories and suggest optimization opportunities?" +``` + +## Troubleshooting + +### Common Issues + +**Agent not responding**: +- Check your internet connection +- Verify the agent is properly configured +- Try refreshing the page +- Contact administrator if issues persist + +**File upload fails**: +- Check file size (may exceed limit) +- Verify file format is supported +- Ensure stable internet connection +- Try uploading smaller files + +**Slow responses**: +- Model may be processing complex request +- High system load may cause delays +- Streaming mode can improve perceived speed +- Break complex questions into smaller parts + +**Context lost**: +- Start a new chat for fresh context +- Avoid extremely long conversations (>20-30 turns) +- Provide necessary context in each message +- Re-upload files if needed in new conversations + +## Security Considerations + +### Data Privacy + +When chatting with agents: +- **PII Detection**: Agents may automatically detect and handle personal information +- **Prompt Injection**: Protection against malicious prompt manipulation +- **Access Logs**: Conversations may be logged for audit purposes +- **Organization Visibility**: Shared agents may be accessible to other users + +### Safe Practices + +- **Don't share passwords** or API keys in chats +- **Mask sensitive data** when providing examples +- **Review agent permissions** before uploading confidential files +- **Use private agents** for sensitive organizational information + +## Next Steps + +Now that you know how to chat with agents: + +1. **[Build Your Own Agent](/agent-forge/using-agent-forge/building-agents)** - Create specialized agents for your needs +2. **[Set Up Knowledge Bases](/agent-forge/using-agent-forge/knowledge-base)** - Enhance agents with organizational knowledge +3. **[Explore Advanced Features](/agent-forge/using-agent-forge/building-agents#generation-settings)** - Fine-tune agent configurations \ No newline at end of file diff --git a/fern/docs/pages/agent-forge/getting-started.mdx b/fern/docs/pages/agent-forge/getting-started.mdx new file mode 100644 index 0000000..2d68c6a --- /dev/null +++ b/fern/docs/pages/agent-forge/getting-started.mdx @@ -0,0 +1,156 @@ +--- +title: Getting Started with Agent Forge +subtitle: This guide walks you through the initial setup process for Agent Forge, from receiving your invitation to logging in and accessing your first agents. +description: Complete guide to creating your account, logging in, and getting started with Agent Forge +--- + +## Prerequisites + +Before you begin, ensure: +- Your organization has a Prediction Guard deployment with Agent Forge enabled +- An administrator has access to invite users to Agent Forge +- You have a valid email address for account creation + +## Step 1: Receiving Your Invitation + +Agent Forge uses an invitation-based system to ensure secure access control. + +### Email Invitation + +Your organization's administrator will send you an invitation email: + +![Email Invitation](./screenshots/00_01_email_invite.png) + +The invitation email contains: +- **From**: Prediction Guard Team +- **Subject**: "You've been invited to Agent Forge!" +- **Content**: Welcome message and acceptance link + +### Accepting the Invitation + +1. **Check your inbox** for the invitation email +2. **Click "Accept Your Invitation"** link in the email +3. You'll be redirected to the registration page + +## User Roles + +Your role is assigned by your organization's administrator at the time of invitation. It determines what you can see and do within Agent Forge. + +| Role | Capabilities | +|------|-------------| +| **Admin** | Invite and remove Agent Forge users, plus all Builder capabilities | +| **Builder** | Create, configure, and manage agents | +| **User** | Chat with agents shared by the organization | + +## Step 2: Creating Your Account + +After clicking the invitation link, you'll be directed to the sign-up page. + +![Sign Up Page](./screenshots/00_02_sign_up.png) + +### Registration Form + +Complete the following fields: + +1. **Name**: Your full name (will be visible to other users in your organization) +2. **Email**: Pre-filled from your invitation (typically your work email) +3. **Password**: Create a strong password +4. **Confirm Password**: Re-enter your password to confirm + +### Completing Registration + +1. **Fill in all required fields** (marked with asterisk *) +2. **Click "Create Account"** button +3. Wait for account creation confirmation +4. You'll be automatically redirected to the login page + +## Step 3: Logging In + +Access Agent Forge through your organization's dedicated URL. + +![Login Page](./screenshots/00_03_login_01.png) + +### Login Process + +1. **Navigate** to `studio.predictionguard.com` (or your organization's custom domain) +2. **Enter your email address** (the one used during registration) +3. **Enter your password** +4. **Click "Log In"** or press Enter + +### First-Time Login + +On your first login, you may be prompted to: +- Complete your profile information +- Accept terms of service +- Set up additional security preferences + +![Login Screen](./screenshots/00_04_login_02.png) + +## Step 4: Password Management + +### Forgot Password + +If you forget your password: + +1. **Click "Forgot Password?"** on the login page +2. **Enter your email address** +3. **Check your inbox** for password reset instructions +4. **Click the reset link** in the email +5. **Create a new password** +6. **Confirm** your new password +7. **Log in** with your new credentials + +## Post-Login: Your First Session + +After successfully logging in, you'll land on the Agent Forge home screen. + +### What You'll See + +- **Sidebar Navigation**: Access to Agent Builder, chats, and knowledge bases +- **Shared Agents**: Agents shared by your organization +- **Your Agents**: Agents you've created (empty initially) +- **Chat Interface**: Ready to start conversations + +### Exploring the Interface + +The Agent Forge interface consists of several key areas: + +**Left Sidebar**: +- **Agent Builder**: Create and manage your agents +- **New Chat**: Start a conversation with an agent +- **Search Chats**: Find previous conversations + +**Main Area**: +- **Agent Selection**: Choose which agent to chat with +- **Chat Interface**: Conversation area +- **Agent Cards**: Browse available agents + +![First Session](./screenshots/01_00_01_First_Session.png) + +## Troubleshooting + +### Common Issues + +**Issue: Invitation link has expired** +- **Solution**: Request a new invitation from your administrator. + +**Issue: Can't create account** +- **Solution**: Verify your email is correct and matches the invitation. Check password requirements. + +**Issue: Login fails with correct credentials** +- **Solution**: Try resetting your password. Ensure Caps Lock is off. Clear browser cache and cookies. + +**Issue: Not seeing any agents** +- **Solution**: Your organization may not have created shared agents yet. Contact your admin or create your own agent. + +## Next Steps + +Now that you're logged in: + +1. **[Explore the Chat Interface](/agent-forge/using-agent-forge/chatting-with-agents)** - Learn how to interact with agents +2. **[Build Your First Agent](/agent-forge/using-agent-forge/building-agents)** - Create a custom agent +3. **[Create a Knowledge Base](/agent-forge/using-agent-forge/knowledge-base)** - Add documents to enhance agent responses + +--- + +**Ready to start chatting?** Continue to [Chatting with Agents](/agent-forge/using-agent-forge/chatting-with-agents) to learn how to interact with AI agents. \ No newline at end of file diff --git a/fern/docs/pages/agent-forge/knowledge-base.mdx b/fern/docs/pages/agent-forge/knowledge-base.mdx new file mode 100644 index 0000000..da85c2f --- /dev/null +++ b/fern/docs/pages/agent-forge/knowledge-base.mdx @@ -0,0 +1,948 @@ +--- +title: Knowledge Base Management +subtitle: Knowledge bases are the foundation of intelligent, context-aware agents in Agent Forge. They enable your agents to access organizational knowledge, documentation, and domain-specific information to provide accurate, informed responses. +description: Complete guide to creating, uploading documents, and managing knowledge bases in Agent Forge +--- + +## What is a Knowledge Base? + +A knowledge base is a collection of documents that agents can search and reference when answering questions. When a user asks a question: + +1. **Semantic Search**: The system searches the knowledge base for relevant content +2. **Retrieval**: Top matching document chunks are retrieved +3. **Context Injection**: Relevant content is added to the agent's context +4. **Response Generation**: Agent generates response using both its training and the retrieved knowledge + +### Benefits of Knowledge Bases + +**Accuracy**: Agents provide information based on your organization's actual documents and policies + +**Currency**: Update knowledge by adding new documents - no need to retrain models + +**Source Attribution**: Agents can cite specific documents or sources + +**Consistency**: All agents using the same knowledge base provide consistent information + +**Specialization**: Create domain-specific knowledge bases for different agent types + +## Accessing Knowledge Base Management + +Navigate to the Knowledge section from the left sidebar or top navigation. + +![Knowledge Base Navigation](./screenshots/04-01-Knowledge-base-section.png) + +The Knowledge Base interface provides: +- **Tabs**: Switch between Agents and Knowledge sections +- **Knowledge Base List**: View all your organization's knowledge bases +- **Create Button**: Create new knowledge bases +- **Search**: Find specific knowledge bases + +## Knowledge Base Overview + +![Knowledge Base Screen](./screenshots/04-02-Knowledge-base-screen.png) + +The main Knowledge Base screen displays: + +**Knowledge Base Cards**: Each showing: +- **Name**: Descriptive title +- **Document Count**: Number of documents in the KB +- **Creation Date**: When the KB was created +- **Edit/Delete Options**: Manage the knowledge base +- **Document Thumbnails**: Preview of included files + +**Example Knowledge Bases**: +- **Generic Knowledge Base**: General organizational information +- **On-Premise Manufacturing Predictive**: Manufacturing-specific knowledge +- **Self Hosting PostgreSQL**: Database documentation +- **AI Agents Data Compliance**: Compliance and regulatory documents + +## Creating a Knowledge Base + +### Step 1: Initiate Creation + +![Create New Knowledge Base](./screenshots/04-03-01-Creating-kb_Create-New-KB.png) + +1. **Click "Create New Knowledge Base"** button in the top-right corner +2. The knowledge base creation form appears + +### Step 2: Basic Information + +**Knowledge Base Name** (Required) +- Descriptive, clear name indicating the KB's content +- Examples: "OWASP Guidelines", "Product Documentation", "Company Policies" + +**Description** +- Explain what documents are included +- Indicate which agents should use this KB +- Note any specific focus areas or limitations + +**Example**: +``` +OWASP Guideline + +This knowledge base contains comprehensive OWASP security +guidelines and best practices for secure application development. +Use with security-focused agents for vulnerability assessment +and secure coding recommendations. +``` + +### Step 3: Advanced Settings + + +**Important**: These settings **cannot be changed** after the knowledge base is created. Plan carefully before proceeding. + + +#### Embedding Model + +**Purpose**: The model used to generate vector embeddings for semantic search + +**Available Models**: +- **bge-m3**: Multilingual, high-quality embeddings (recommended) +- **bridgetower**: Multimodal embeddings (text + images) +- Other models based on your deployment + +**Choosing an Embedding Model**: + +| Model | Best For | Languages | Notes | +|-------|----------|-----------|-------| +| bge-m3 | General purpose, multilingual | 100+ languages | Excellent balance | +| bridgetower | Documents with images | English | Multimodal support | + +**Recommendation**: Use **bge-m3** for most use cases unless you have specific requirements. + + +**Why It Matters**: The embedding model must remain consistent for a knowledge base. Changing it would require regenerating all embeddings, which is why it's locked after creation. + + +#### Chunk Size + +**Range**: 100-8192 characters + +**Default**: 1500 characters + +**Purpose**: Size of text segments for indexing and retrieval + +**How It Works**: +- Documents are split into chunks of approximately this size +- Each chunk is embedded separately +- During search, individual chunks are matched and retrieved +- Multiple chunks can be retrieved from the same document + +**Choosing Chunk Size**: + +| Size | Use Case | Advantages | Disadvantages | +|------|----------|------------|---------------| +| 500-1000 | Short Q&A, definitions | Precise matching | May miss context | +| 1000-2000 | General purpose (recommended) | Good balance | Standard | +| 2000-4000 | Long-form content | More context per chunk | Less precise | +| 4000-8192 | Technical docs, code | Maximum context | May include irrelevant info | + +**Best Practices**: +``` +FAQ documents: 800-1200 characters +Technical manuals: 1500-2000 characters +Legal documents: 2000-3000 characters +Code repositories: 1500-2500 characters +``` + +**Example**: With chunk size 1500: +- 3000-word document = ~6-8 chunks +- Each chunk contains 200-250 words +- Chunks overlap to maintain context + +#### Chunk Overlap + +**Range**: 0-500+ characters + +**Default**: 100 characters + +**Purpose**: Number of characters that overlap between consecutive chunks + +**Why Overlap Matters**: +- Prevents breaking related information across chunks +- Maintains context at chunk boundaries +- Improves retrieval quality + +**Visual Example**: +``` +Chunk Size: 1500, Overlap: 100 + +Document: [------------------------3000 characters------------------------] + +Chunk 1: [-------1500-------] + [100] <- Overlap +Chunk 2: [-------1500-------] + [100] <- Overlap +Chunk 3: [-------1500-------] +``` + +**Recommended Overlap**: +``` +Chunk 500: Overlap 50 (10%) +Chunk 1500: Overlap 100-150 (7-10%) +Chunk 3000: Overlap 200-300 (7-10%) +``` + +**Higher Overlap**: +- ✅ Better context preservation +- ✅ Fewer missed connections +- ❌ More storage required +- ❌ Potential redundancy + +**Lower Overlap**: +- ✅ More efficient storage +- ✅ Faster processing +- ❌ May break related content +- ❌ Context gaps at boundaries + + +**Tip**: For most use cases, set overlap to 7-10% of chunk size. This provides good context preservation without excessive redundancy. + + +### Step 4: Create the Knowledge Base + +1. **Review all settings** carefully (Advanced Settings are permanent) +2. **Click "Create Knowledge Base"** +3. Empty knowledge base is created +4. You'll be directed to the document upload interface + +## Uploading Documents + +After creating a knowledge base, you can add documents. + +![Upload Documents](./screenshots/04-03-02-Creating-kb_Upload-Files.png) + +### Supported File Formats + +Agent Forge supports a wide range of document formats: + +**Document Formats**: +- `.pdf` - PDF documents +- `.doc` - Microsoft Word (legacy) +- `.docx` - Microsoft Word +- `.txt` - Plain text files +- `.md` - Markdown files + +**Structured Data**: +- `.csv` - Comma-separated values +- `.json` - JSON data files + +**Other Formats** (check your deployment): +- `.pptx` - PowerPoint presentations +- `.xlsx` - Excel spreadsheets +- `.html` - Web pages +- `.xml` - XML documents + +### Document Management + +Once uploaded, each document shows: +- **Filename**: Original document name +- **Status**: Processing status badge +- **Toggle**: Enable/disable document +- **Delete**: Remove document from knowledge base + +**Enable/Disable Documents**: +- Use the toggle switch to enable or disable documents +- **Disabled documents** are not searched +- Useful for temporarily excluding documents without deleting + + +**Note**: Only **enabled** documents will be used for AI retrieval. Disable documents you want to keep but not search. + + +## Testing Semantic Search + +One of the most powerful features of Agent Forge is the ability to test your knowledge base's semantic search directly - before attaching it to any agents. This helps you verify that your documents are properly indexed and that the search is returning relevant results. + +### Accessing the Preview Feature + +To test semantic search on a knowledge base: + +1. **Navigate to the Knowledge section** from the sidebar +2. **Click on a knowledge base** card to open it +3. **Click the "Preview" button** at the top of the knowledge base detail page +4. The semantic search testing interface will appear + +![Test Semantic Search](./screenshots/04-03-03-Test_sim-matching_kb.png) + +### Understanding the Preview Interface + +The Preview interface provides a complete testing environment for your knowledge base: + +**Header Description**: +``` +"Test similarity matching on all enabled documents in the knowledge base. +This will return a list of the top 7 matching document chunks." +``` + +This feature searches across **all enabled documents** in your knowledge base using the same semantic search technology that agents use. + +### Search Components + +#### Search Query Box + +**Purpose**: Enter your test query to search the knowledge base + +**Character Limit**: 2000 characters + +**What to Test**: +- Natural language questions +- Keywords and phrases +- Specific concepts or terms +- Edge cases and variations + + +**Tip**: Use the same types of questions your users will ask agents. This gives you realistic preview of search performance. + + +#### Maximum Results Slider + +**Range**: 1 to 20 chunks + +**Purpose**: Control how many document chunks are retrieved + +**Default**: Typically set to 7 (as shown in interface) + +**How to Use**: +- **Lower values (1-5)**: Test if top results are highly relevant +- **Medium values (5-10)**: Standard testing (balanced) +- **Higher values (10-20)**: See breadth of relevant content + +**What to Look For**: +- Are the top results most relevant? +- At what point do results become less relevant? +- Do you have enough relevant chunks for complex queries? + +#### Minimum Relevance Score Slider + +**Range**: Looser (0.0) to Stricter (1.0) + +**Purpose**: Filter results by similarity score threshold + +**How It Works**: +- Each result has a similarity score from 0 to 1 +- 0 = No similarity, 1 = Perfect match +- Only results above the threshold are returned + +**Slider Positions**: + +| Position | Score Range | Use Case | +|----------|-------------|----------| +| Looser (left) | 0.3-0.5 | Broad, exploratory searches | +| Middle | 0.5-0.7 | Balanced (recommended for testing) | +| Stricter (right) | 0.7-1.0 | High precision requirements | + +**Testing Strategy**: +1. Start with **middle position** +2. Run your test query +3. If too many irrelevant results: **Move right** (stricter) +4. If too few results: **Move left** (looser) +5. Find the sweet spot for your content + +### Understanding Search Results + +After clicking **Submit**, you'll see results displayed with detailed information: + +#### Results Header + +``` +Results (7) +``` + +Shows the total number of chunks returned matching your criteria. + +#### Individual Result Cards + +Each result displays comprehensive information: + +**Document Name**: +``` +OWASP-Top-10-for-LLMs-2023-slides-v1_0_copy.pdf +``` +The source document containing this chunk. + +**Chunk Index**: +``` +Chunk Index: 4 +``` +The position of this chunk within the document (0-indexed). Useful for understanding document structure. + +**Similarity Score**: +``` +Similarity Score (0-1): 0.57 +``` +How closely this chunk matches your query: +- **0.8-1.0**: Highly relevant, near-exact match +- **0.6-0.8**: Good relevance, strong match +- **0.4-0.6**: Moderate relevance, partial match +- **0.2-0.4**: Weak relevance, tangential +- **0.0-0.2**: Very weak, likely irrelevant + +**Chunk Preview**: +``` +ATTACK SCENARIOS S An attacker provides a +direct prompt injection to an LLM-based support +chatbot ATTACK SCENARIOS S An attacker... +``` +First ~150-200 characters of the chunk content. Lets you verify relevance at a glance. + +**View All Link**: +Click to see the complete chunk text (up to your configured chunk size). + +### Interpreting Results + +#### What Good Results Look Like + +**High Relevance Scores**: +- Top 3-5 results should have scores **above 0.6** +- At least one result should be **above 0.7** for specific queries + +**Appropriate Content**: +- Chunks directly address the query +- Information is accurate and complete +- Context is preserved (no mid-sentence cuts) + +**Good Distribution**: +- Results come from relevant documents +- Multiple perspectives if applicable +- Logical ordering by relevance + +**Example Good Result**: +``` +Query: "prompt injection attacks" + +Result #1: +Document: OWASP-Top-10-for-LLMs-2023.pdf +Chunk Index: 12 +Similarity Score: 0.87 +Preview: "Prompt injection attacks occur when an attacker +manipulates the input to an LLM in order to override its +original instructions..." +✅ Highly relevant, directly addresses query +``` + +#### Warning Signs + +**Low Relevance Scores**: +``` +All results below 0.5 = Potential issues: +- Documents don't contain relevant information +- Chunk size may be too large/small +- Query phrasing doesn't match document language +``` + +**Irrelevant Content**: +``` +Query: "authentication best practices" +Result: Content about "database optimization" +Similarity Score: 0.45 + +❌ Not relevant - may indicate: +- Missing content in knowledge base +- Poor document organization +- Need for additional documents +``` + +**Broken Context**: +``` +Preview: "...the process requires that before starting +you must ensure..." + +❌ Mid-sentence start, missing context +- May need increased chunk overlap +- Document structure issues +``` + +### Testing Strategies + +#### 1. Comprehensive Coverage Testing + +**Goal**: Verify knowledge base covers intended topics + +**Process**: +``` +1. List 10-15 key topics your KB should cover +2. Create test queries for each topic +3. Run searches and check results +4. Identify gaps in coverage +5. Add missing documents +``` + +**Example Test Plan**: +``` +Knowledge Base: "Security Best Practices" + +Test Queries: +✓ "SQL injection prevention" → Good results (0.78) +✓ "XSS mitigation" → Good results (0.82) +✗ "CSRF protection" → No relevant results + → Action: Add CSRF documentation +✓ "Authentication methods" → Good results (0.71) +✗ "Session management" → Low scores (0.43) + → Action: Improve existing docs or add content +``` + +#### 2. Relevance Threshold Testing + +**Goal**: Find optimal minimum relevance score for your KB + +**Process**: +``` +1. Choose 5 representative test queries +2. Start with score at "Looser" (left) +3. Run each query, note result quality +4. Move slider right incrementally +5. Find sweet spot where: + - Good results are included + - Irrelevant results are filtered out +``` + +**Documentation**: +``` +Query: "GDPR compliance requirements" + +Looser (0.3): 15 results, 8 relevant, 7 irrelevant +Medium (0.5): 9 results, 8 relevant, 1 irrelevant ✓ +Stricter (0.7): 4 results, 4 relevant, 0 irrelevant ✓ + +Recommendation: Set agent minimum score to 0.5-0.7 +``` + +#### 3. Chunk Size Validation + +**Goal**: Verify chunk size settings are appropriate + +**Process**: +``` +1. Run several test queries +2. Click "View all" on results +3. Check if chunks are: + - Complete thoughts/paragraphs + - Not cut off mid-sentence + - Contain sufficient context + - Not overly long with irrelevant info +``` + +**If chunks are problematic**: +- **Too small** (broken context): Create new KB with larger chunk size +- **Too large** (diluted relevance): Create new KB with smaller chunk size +- **Poor boundaries**: Increase chunk overlap in new KB + + +**Remember**: Chunk size and overlap cannot be changed after KB creation. Create a new KB if needed. + + +#### 4. Query Variation Testing + +**Goal**: Ensure KB works with different phrasings + +**Process**: +``` +Test the same concept with different queries: + +Query 1: "How do I prevent SQL injection?" +Query 2: "SQL injection protection methods" +Query 3: "Defending against SQL attacks" +Query 4: "Database input sanitization" + +All should return similar, relevant results. +If not → May need more diverse documents or better document coverage +``` + +### Best Practices for Testing + +**Before Attaching to Agents**: +1. ✅ Test with 20-30 representative queries +2. ✅ Verify top results have scores above 0.6 +3. ✅ Check that chunks contain complete, useful information +4. ✅ Confirm coverage of all intended topics +5. ✅ Test edge cases and unexpected queries +6. ✅ Document optimal relevance threshold + +**Regular Testing**: +- **After adding documents**: Test that new content is retrievable +- **Monthly**: Run standard test queries to ensure consistency +- **After changes**: Verify updates haven't broken existing searches + +**Document Your Findings**: +``` +Knowledge Base: Security Guidelines +Tested: 2024-02-14 +Optimal Settings: + - Minimum Relevance Score: 0.55-0.65 + - Recommended Max Results: 8-10 +Coverage: + ✓ OWASP Top 10 + ✓ Authentication + ✓ Encryption + ✗ API Security (needs addition) +Action Items: + - Add API security documentation + - Update outdated sections on TLS +``` + +### Troubleshooting Search Issues + +**Issue: No results returned** +- **Check**: Are documents enabled? +- **Check**: Is minimum relevance score too strict? +- **Try**: Move slider to "Looser" +- **Try**: Rephrase query with simpler terms +- **Solution**: May need to add content on this topic + +**Issue: Low relevance scores across the board** +- **Check**: Do documents actually cover this topic? +- **Check**: Is query phrasing very different from document language? +- **Try**: Use terminology from your documents +- **Solution**: May need to add or improve documents + +**Issue: Too many irrelevant results** +- **Check**: Is minimum relevance score too loose? +- **Try**: Move slider to "Stricter" +- **Try**: Be more specific in query +- **Solution**: Consider removing off-topic documents + +**Issue: Results are out of order by relevance** +- **Check**: Scores are very close (e.g., 0.67 vs 0.66) +- **This is normal**: Slight variations in scoring +- **Solution**: If top results are relevant, this is fine + +### Using Test Results to Configure Agents + +After thorough testing, use your findings to configure agents: + +**In Agent Builder** → **Resources & Knowledge** → **Customize Search Settings**: + +1. **Minimum Relevance Score**: Set based on your testing + ``` + If testing showed 0.55 works well → Set agent to 0.5-0.6 + ``` + +2. **Maximum Results**: Based on how many chunks are typically relevant + ``` + If top 5-8 results usually relevant → Set to 8-10 + ``` + +3. **Maximum Context Size**: Based on chunk sizes and result counts + ``` + 10 results × 1500 chars/chunk = 15,000 minimum context size + ``` + +### Example Testing Workflow + +**Complete Example**: Testing a new Security KB + +``` +Step 1: Initial Upload +- Uploaded 15 security documents +- All processing completed + +Step 2: Coverage Testing +Query: "SQL injection" → 0.82 ✓ +Query: "XSS prevention" → 0.79 ✓ +Query: "CSRF attacks" → 0.34 ✗ (needs content) +Query: "Authentication" → 0.71 ✓ +Query: "Session security" → 0.48 ⚠️ (weak) + +Step 3: Threshold Testing +Tested at 0.4, 0.5, 0.6, 0.7 +Optimal: 0.55 (good balance) + +Step 4: Chunk Quality Check +Viewed full chunks - mostly complete +Some cut mid-paragraph (acceptable) + +Step 5: Recommendations +- Add CSRF documentation (missing) +- Enhance session security content (weak) +- Set agent minimum relevance: 0.55 +- Set agent max results: 8 + +Step 6: Actions Taken +- Added CSRF guide document +- Re-tested: now 0.73 ✓ +- Configured security agent with findings +- Documented for team +``` + +### Next Steps After Testing + +Once you're satisfied with test results: + +1. **[Configure Agents](/agent-forge/using-agent-forge/building-agents#resources--knowledge)** - Attach KB to agents with optimal settings +2. **Monitor Performance** - Track agent response quality +3. **Iterate** - Continuously improve based on user feedback +4. **Document** - Keep notes on what works best + + +**Pro Tip**: Bookmark high-performing test queries and run them periodically to ensure consistent KB quality over time. + + +## Best Practices for Knowledge Bases + +### Document Preparation + +**Before Uploading**: + +1. **Clean Documents**: + - Remove unnecessary headers/footers + - Fix formatting issues + - Remove duplicate content + - Ensure text is selectable (not scanned images without OCR) + +2. **Organize Content**: + - Use clear headings and structure + - Break long documents into logical sections + - Include table of contents for long docs + - Use consistent formatting + +3. **Add Metadata**: + - Include document titles + - Add creation/update dates + - Note document version if applicable + - Include author or source information + +**Document Quality**: +``` +✅ Well-structured PDFs with clear headings +✅ Updated, current information +✅ Accurate, fact-checked content +✅ Clear, professional writing + +❌ Scanned images without text layer +❌ Outdated, deprecated information +❌ Duplicate or redundant content +❌ Poor quality or corrupted files +``` + +### Organizing Knowledge Bases + +**Strategy 1: By Domain** +- "Security Knowledge Base" - All security-related docs +- "Compliance Knowledge Base" - Regulatory and compliance docs +- "Product Documentation" - Product guides and manuals + +**Strategy 2: By Audience** +- "Customer Facing KB" - Public information +- "Internal Operations KB" - Internal procedures +- "Technical Team KB" - Developer documentation + +**Strategy 3: By Update Frequency** +- "Static Policies KB" - Rarely changing policies +- "Product Updates KB" - Frequently updated features +- "Current Events KB" - Time-sensitive information + +### Maintenance + +**Regular Reviews**: +- **Audit content** quarterly +- **Remove outdated** documents +- **Update changed** information +- **Add new** relevant documents + +**Quality Checks**: +- Test knowledge base with sample queries +- Verify agents are retrieving correct information +- Check for gaps in coverage +- Monitor user feedback on agent responses + +## Editing Knowledge Bases + +To edit an existing knowledge base: + +1. **Navigate to Knowledge section** +2. **Click on the knowledge base** card +3. **Edit Basic Information**: + - Update name + - Modify description +4. **Manage Documents**: + - Upload additional documents + - Enable/disable existing documents + - Delete unwanted documents +5. **Click "Update Knowledge Base"** + + +**Remember**: Advanced Settings (embedding model, chunk size, chunk overlap) **cannot be changed** after creation. + + +## Attaching Knowledge Bases to Agents + +Once created, attach knowledge bases to agents: + +1. **Go to Agent Builder** +2. **Create or edit an agent** +3. **Navigate to "Resources & Knowledge" section** +4. **Select "Knowledge Bases"** dropdown +5. **Choose one or more knowledge bases** +6. **Configure search settings** (optional) +7. **Save agent** + +See [Building Agents - Resources & Knowledge](/agent-forge/using-agent-forge/building-agents#resources--knowledge) for detailed instructions. + +## Testing Knowledge Bases + +### Verification Steps + +After uploading documents: + +1. **Create a test agent** with the knowledge base attached +2. **Ask questions** you know are answered in the documents +3. **Verify responses** cite correct information +4. **Test edge cases** - questions with no answer in KB +5. **Check relevance** - are retrieved chunks actually relevant? + +### Sample Test Queries + +**Direct Fact Retrieval**: +``` +"What is the maximum file size allowed?" +Expected: Agent retrieves and cites exact limit from docs +``` + +**Conceptual Questions**: +``` +"Explain the process for creating a new agent" +Expected: Agent synthesizes information from relevant sections +``` + +**Multi-Document Queries**: +``` +"What are the differences between security settings and access control?" +Expected: Agent combines information from multiple docs +``` + +**Not in Knowledge Base**: +``` +"What's the weather today?" +Expected: Agent acknowledges this isn't in its knowledge base +``` + +## Troubleshooting + +### Common Issues + +**Issue: Agent not using knowledge base** +- **Solution**: + - Verify KB is attached to agent + - Check documents are enabled + - Lower minimum relevance score + - Increase maximum results + +**Issue: Irrelevant results retrieved** +- **Solution**: + - Increase minimum relevance score + - Reduce chunk size for more precise matching + - Review document quality + - Remove off-topic documents + +**Issue: Document upload fails** +- **Solution**: + - Check file size (under 50MB) + - Verify file format is supported + - Ensure file isn't corrupted + - Try uploading again + +**Issue: Processing takes very long** +- **Solution**: + - Large files take longer to process + - Complex PDFs may need more time + - Check status periodically + - Contact admin if stuck for hours + +**Issue: Chunks don't contain complete information** +- **Solution**: + - Increase chunk size + - Increase chunk overlap + - Restructure source documents for better segmentation + - Note: Can't change existing KB - create new one + +### Performance Optimization + +**Faster Search**: +- Use focused knowledge bases (fewer documents) +- Remove redundant documents +- Optimize chunk size for your use case +- Set appropriate maximum results limit + +**Better Results**: +- Higher quality source documents +- Appropriate chunk size for content type +- Good chunk overlap (7-10%) +- Regular content updates + +## Advanced Topics + +### Semantic Search Explained + +Knowledge bases use **semantic search**, not keyword search: + +**Keyword Search** (Traditional): +``` +Query: "How do I reset my password?" +Matches: Documents containing words "reset" and "password" +``` + +**Semantic Search** (Agent Forge): +``` +Query: "How do I reset my password?" +Matches: Documents about: +- Password recovery +- Account access restoration +- Credential reset procedures +Even if they don't use exact words "reset password" +``` + +### Embedding Models + +Embeddings convert text into numerical vectors that capture meaning: + +``` +"The cat sat on the mat" → [0.234, -0.456, 0.789, ...] +"A feline rested on the rug" → [0.221, -0.442, 0.801, ...] +``` + +Similar meanings produce similar vectors, enabling semantic search. + +### Vector Databases + +Knowledge bases use vector databases (e.g., FAISS, Pinecone, Milvus) to: +- Store embeddings efficiently +- Perform fast similarity search +- Scale to millions of documents +- Enable real-time retrieval + +## Security Considerations + +### Sensitive Information + +**Before uploading**: +- Review documents for sensitive data +- Redact confidential information +- Consider access control implications +- Follow your organization's data policies + +**Remember**: +- Knowledge base content may be visible to all agents using it +- Agents with PII detection can mask sensitive data, but prevention is better +- Audit who has access to agents using sensitive KBs + +### Access Control + +Currently, knowledge bases are organization-wide resources: +- Any agent creator can attach any KB +- Users chatting with agents see information from attached KBs +- Plan knowledge base scope accordingly + +**Best Practices**: +- Create separate KBs for different sensitivity levels +- Use clear naming to indicate confidentiality +- Document which agents should use which KBs +- Regular access audits + +## Next Steps + +Now that you understand knowledge bases: + +1. **[Build Agents](/agent-forge/using-agent-forge/building-agents)** - Create agents that use your knowledge bases +2. **[Chat with Agents](/agent-forge/using-agent-forge/chatting-with-agents)** - Test agents with knowledge base integration +3. **Iterate and Improve** - Refine knowledge bases based on agent performance \ No newline at end of file diff --git a/fern/docs/pages/agent-forge/mcp.mdx b/fern/docs/pages/agent-forge/mcp.mdx new file mode 100644 index 0000000..bf65e0f --- /dev/null +++ b/fern/docs/pages/agent-forge/mcp.mdx @@ -0,0 +1,122 @@ +--- +title: MCP (Model Context Protocol) +subtitle: Agent Forge includes native support for the **Model Context Protocol (MCP)** — an open standard that allows AI agents to securely connect to external tools, APIs, databases, and services. With MCP, your agents can go beyond language generation to take real actions (querying live data, executing code, calling external APIs, and interacting with organizational systems). +description: Learn how to configure organizational and agent-level MCP servers in Agent Forge to extend agent capabilities +--- + +## What is MCP? + +MCP (Model Context Protocol) is a standardized protocol for connecting AI models to external capabilities through **MCP servers**. Each MCP server exposes a set of tools that the agent can invoke during a conversation. + +**Examples of what MCP enables**: +- Query a live database or data warehouse +- Retrieve real-time information from external APIs +- Execute code or automation scripts +- Read from and write to organizational systems (CRMs, ticketing, file stores) +- Interact with third-party services and integrations + +## MCP Levels in Agent Forge + +Agent Forge supports MCP at two levels: + +| Level | Configured By | Available To | +|-------|--------------|--------------| +| **Organization-level** | Admin via Unified CX (coming soon!) | All agents in the organization | +| **Agent-level** | Agent Builder and above | That specific agent only | + +This layered approach lets administrators centrally manage shared infrastructure while allowing individual agent builders to extend with specialized tools. + +## Configuring MCP in the Agent Builder + +MCP is configured during agent creation or editing, within the **Agent Builder**. + +### Step 1: Enable "Non-Streaming (With MCP Capabilities)" Mode + +![Agent Builder - Select Streaming Mode](./screenshots/05-01-01-MCP-AgentBuilder-Select-Streaming-Mode.png) + +The MCP support requires **Non-Streaming (With MCP Capabilities)** mode to be enabled. + +### Step 2: Select an Organization-Level MCP Server (If Available) + +![Agent Builder - Select Org-Level MCP](./screenshots/05-01-02-MCP-AgentBuilder-Select-OrgLevelMCP_IfAvailable.png) + +If your organization's administrator has pre-configured MCP servers, they will appear in the **MCP Tools Available in Your Organization** section of the Agent Builder. These are shared, centrally-managed servers available to all agent builders in your organization. + + +**Tip**: Organization-level MCP servers are ideal for unified governance and shared infrastructure — internal databases, company APIs, and standard integrations that multiple agents can access. + + +### Step 3: Add a Custom MCP Server (Agent-Level) + +![Agent Builder - Add Custom MCP](./screenshots/05-01-03-MCP-AgentBuilder-AddCustomMCP.png) + +In addition to organization-level servers, you can connect your agent to a **custom MCP server** specific to your agent's needs. + +**To add a custom MCP server**: + +1. In the **MCP Server Configuration (JSON)** field, enter your server config as a JSON array +2. Each entry in the array defines one MCP server — add multiple objects to connect multiple servers +3. Click **Save** to attach the configuration to your agent + +**Configuration format**: + +```json +[ + { + "type": "mcp", + "server_url": "https://your-mcp-server.yourcompany.com/mcp/", + "server_label": "your-server-label", + "server_description": "Description of what this MCP server provides", + "authorization": "Your Bearer Token", + "allowed_tools": ["tool_name_1", "tool_name_2"] + } +] +``` + +| Field | Description | +|-------|-------------| +| `type` | Always `"mcp"` | +| `server_url` | The full URL of your MCP server endpoint | +| `server_label` | A short identifier for the server (no spaces) | +| `server_description` | Human-readable description of the server's capabilities | +| `authorization` | Bearer token or other auth credential for the server | +| `allowed_tools` | List of specific tool names to expose to the agent | + +## How Agents Use MCP Tools + +Once MCP servers are configured, the agent automatically has access to the tools exposed by those servers. During a conversation: + +1. **User sends a message** requiring external data or action +2. **Agent determines** which MCP tool(s) to invoke +3. **MCP tool executes** against the connected server +4. **Results are returned** to the agent +5. **Agent incorporates results** into its streamed response + +This all happens transparently — users see a natural conversational response that incorporates live data. + +## Use Cases + +### Data Retrieval +Connect agents to internal databases, data warehouses, or BI tools so they can query live figures, metrics, and reports on demand. + +### System Integration +Enable agents to interact with ticketing systems, CRMs, or internal tools — looking up records, creating entries, or updating statuses. + +### Live API Access +Give agents access to external APIs (weather, financial data, inventory) so responses reflect current real-world state, not just training knowledge. + +### Code Execution +Attach MCP servers that execute code or scripts, enabling agents to perform calculations, data transformations, or automation tasks. + +## Best Practices + +**Name Servers Descriptively**: When adding custom MCP servers, use clear names that indicate what system or capability they represent (e.g., "Inventory DB", "Jira API", "Analytics Platform"). + +**Test After Configuration**: After saving your agent with MCP configured, start a test conversation and trigger a tool call to verify the MCP connection is working correctly. + +**Limit Tool Scope**: Configure MCP servers to expose only the tools your agent actually needs. A focused set of tools reduces confusion and improves agent decision-making. + +## Next Steps + +- [Building Agents](/agent-forge/using-agent-forge/building-agents) - Full guide to agent configuration +- [Knowledge Base](/agent-forge/using-agent-forge/knowledge-base) - Add RAG-based organizational knowledge alongside MCP tools \ No newline at end of file diff --git a/fern/docs/pages/agent-forge/overview.mdx b/fern/docs/pages/agent-forge/overview.mdx new file mode 100644 index 0000000..4cdc12b --- /dev/null +++ b/fern/docs/pages/agent-forge/overview.mdx @@ -0,0 +1,129 @@ +--- +title: Agent Forge Overview +subtitle: Agent Forge is Prediction Guard's intuitive web-based platform for building, managing, and deploying AI agents without writing code. It enables teams to create specialized AI assistants tailored to specific use cases, complete with knowledge bases, MCP tool integrations, security controls, and custom configurations. +description: Learn about Agent Forge, Prediction Guard's platform for building, managing, and deploying AI agents +--- + +## What is Agent Forge? + +Agent Forge is a comprehensive agent development and management platform that allows you to: + +- **Build Custom AI Agents**: Create specialized agents with custom instructions, knowledge bases, and configurations +- **No-Code Interface**: Design and deploy agents through an intuitive web interface without programming +- **Knowledge Base Integration**: Connect agents to organizational knowledge through document uploads and vector search +- **Enterprise Security**: Built-in PII detection, prompt injection protection, and access controls +- **Collaborative Platform**: Share agents across your organization and manage team access +- **MCP Tool Integration**: Connect agents to external tools, APIs, and data sources via native Model Context Protocol support +- **Production-Ready**: Deploy agents that integrate seamlessly with your Prediction Guard infrastructure + +## Key Capabilities + +### Agent Building & Configuration + +Agent Forge provides comprehensive tools for building sophisticated AI agents: + +- **Model Selection**: Choose from available LLMs deployed in your Prediction Guard instance +- **System Prompts**: Define agent behavior, personality, and expertise with detailed instructions +- **Advanced Generation Settings**: Fine-tune temperature, Top P, Top K, and max output tokens for optimal responses + +### MCP Tool Integration + +Extend agent capabilities beyond language generation with native Model Context Protocol (MCP) support: + +- **Organization-Level MCP**: Administrators can configure shared MCP servers available to all agents +- **Agent-Level MCP**: Individual agents can connect to custom MCP servers for specialized needs +- **Live Data Access**: Query databases, APIs, and external services in real-time + +### Knowledge Base Management + +Empower your agents with organizational knowledge: + +- **Document Upload**: Support for various file formats including PDFs, text files, and more +- **Vector Search**: Automatic embedding generation and semantic search capabilities +- **Advanced Vector Search Settings**: Configure minimum relevance thresholds for knowledge retrieval +- **Context Management**: Control maximum results and context size for optimal performance + +### Roles & Permissions + +Agent Forge uses a role-based access model to control what each user can do: + +- **Admin**: Invite and remove users, plus all Builder capabilities +- **Builder**: Create, configure, and manage agents +- **User**: Chat with agents shared across the organization + +### Security & Access Control + +Enterprise-grade security features made available to every agent: + +- **Prompt Injection Protection**: Detect and block malicious prompt injection attacks +- **PII Detection & Replacement**: Automatically identify and redact personally identifiable information +- **Topic Filtering**: Control conversation scope and prevent off-topic interactions +- **Organization Sharing**: Manage agent visibility and access across your organization + +### Chat Interface + +Intuitive interface for interacting with agents: + +- **Conversation Management**: Create, search, and organize chat sessions +- **File Upload Support**: Share documents with agents during conversations +- **Drag-and-Drop**: Easy file attachment through drag-and-drop interface + +## Use Cases + +Agent Forge is designed for a wide range of enterprise applications: + +### Compliance & Regulatory Agents +Build agents specialized in regulatory compliance, such as financial services compliance monitoring, healthcare data regulations, or industry-specific standards adherence. + +### Security Advisory Agents +Create security-focused agents that help teams navigate security protocols, analyze vulnerabilities, or provide best practices for secure development. + +### Operations Optimization +Deploy agents that assist with operational efficiency, cost analysis, system monitoring, or process automation recommendations. + +### Knowledge Management +Build internal knowledge agents that help employees find information, understand company policies, or navigate complex documentation. + +### Industry-Specific Assistants +Create domain-expert agents for retail operations, manufacturing processes, logistics management, or any specialized industry need. + +## Quick Start + +Getting started with Agent Forge is simple: + +1. **Receive Invitation**: Your organization admin will send you an email invitation +2. **Create Account**: Sign up with your email and create a secure password +3. **Explore Shared Agents**: Start by chatting with agents shared by your organization +4. **Build Your First Agent**: Click "Create New Agent" and configure your custom agent ('Member' level and above) +5. **Connect MCP Tools**: Attach organization or custom MCP servers to extend your agent's capabilities +6. **Add Knowledge**: Upload relevant documents to create a knowledge base +7. **Test & Iterate**: Chat with your agent and refine its configuration +8. **Share with Team**: Enable organization sharing when ready for team use + +## Benefits + +### For Developers +- **Rapid Prototyping**: Build and test agent concepts in minutes +- **No Backend Code**: Focus on agent behavior, not infrastructure +- **API Integration**: Agents use your existing Prediction Guard deployment + +### For Business Users +- **Self-Service**: Create agents without technical dependencies +- **Domain Expertise**: Leverage your knowledge to build specialized assistants +- **Immediate Value**: Deploy production-ready agents quickly + +### For Organizations +- **Centralized Management**: Single platform for all organizational agents +- **Knowledge Sharing**: Distribute expertise through shared agents +- **Security & Compliance**: Enterprise-grade controls and audit trails +- **Cost Efficiency**: Maximize ROI on your Prediction Guard investment + +## Next Steps + +Ready to get started? Continue with: + +- [Getting Started](/agent-forge/getting-started/getting-started) - Account setup and first login +- [Chatting with Agents](/agent-forge/using-agent-forge/chatting-with-agents) - Learn the chat interface +- [Building Agents](/agent-forge/using-agent-forge/building-agents) - Create your first agent +- [MCP Integration](/agent-forge/using-agent-forge/mcp) - Connect agents to external tools and data sources +- [Knowledge Base](/agent-forge/using-agent-forge/knowledge-base) - Add RAG-based knowledge base to agents \ No newline at end of file diff --git a/fern/docs/pages/agent-forge/screenshots/00_01_email_invite.png b/fern/docs/pages/agent-forge/screenshots/00_01_email_invite.png new file mode 100644 index 0000000..dc3a380 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/00_01_email_invite.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/00_02_sign_up.png b/fern/docs/pages/agent-forge/screenshots/00_02_sign_up.png new file mode 100644 index 0000000..8783f7d Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/00_02_sign_up.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/00_03_login_01.png b/fern/docs/pages/agent-forge/screenshots/00_03_login_01.png new file mode 100644 index 0000000..670eb1f Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/00_03_login_01.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/00_04_login_02.png b/fern/docs/pages/agent-forge/screenshots/00_04_login_02.png new file mode 100644 index 0000000..7f1fd9c Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/00_04_login_02.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/01_00_01_First_Session.png b/fern/docs/pages/agent-forge/screenshots/01_00_01_First_Session.png new file mode 100644 index 0000000..0a33392 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/01_00_01_First_Session.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/01_00_chatting_with_agents_first_screen_select_agent.png b/fern/docs/pages/agent-forge/screenshots/01_00_chatting_with_agents_first_screen_select_agent.png new file mode 100644 index 0000000..0cecbc9 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/01_00_chatting_with_agents_first_screen_select_agent.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/01_01_begin_conversation.png b/fern/docs/pages/agent-forge/screenshots/01_01_begin_conversation.png new file mode 100644 index 0000000..84aa659 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/01_01_begin_conversation.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/01_02_chat_management.png b/fern/docs/pages/agent-forge/screenshots/01_02_chat_management.png new file mode 100644 index 0000000..6e0cb45 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/01_02_chat_management.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/02_01_file-upload.png b/fern/docs/pages/agent-forge/screenshots/02_01_file-upload.png new file mode 100644 index 0000000..f03aa07 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/02_01_file-upload.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/02_02_drag-and-drop-file-upload.png b/fern/docs/pages/agent-forge/screenshots/02_02_drag-and-drop-file-upload.png new file mode 100644 index 0000000..09e6b9d Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/02_02_drag-and-drop-file-upload.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/03-01-Agent-builder-screen.png b/fern/docs/pages/agent-forge/screenshots/03-01-Agent-builder-screen.png new file mode 100644 index 0000000..324ed40 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/03-01-Agent-builder-screen.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/03-02-01-Creating-agent_Create-new-agent-button.png b/fern/docs/pages/agent-forge/screenshots/03-02-01-Creating-agent_Create-new-agent-button.png new file mode 100644 index 0000000..74f4cfc Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/03-02-01-Creating-agent_Create-new-agent-button.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/03-02-02-Creating-agent_Basic-info-sys-prompt-streaming-mode.jpg b/fern/docs/pages/agent-forge/screenshots/03-02-02-Creating-agent_Basic-info-sys-prompt-streaming-mode.jpg new file mode 100644 index 0000000..e40b175 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/03-02-02-Creating-agent_Basic-info-sys-prompt-streaming-mode.jpg differ diff --git a/fern/docs/pages/agent-forge/screenshots/03-02-02-Creating-agent_Generation-settings.jpg b/fern/docs/pages/agent-forge/screenshots/03-02-02-Creating-agent_Generation-settings.jpg new file mode 100644 index 0000000..074e169 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/03-02-02-Creating-agent_Generation-settings.jpg differ diff --git a/fern/docs/pages/agent-forge/screenshots/03-02-03-Creating-agent_Knowledge-base.png b/fern/docs/pages/agent-forge/screenshots/03-02-03-Creating-agent_Knowledge-base.png new file mode 100644 index 0000000..1419dd4 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/03-02-03-Creating-agent_Knowledge-base.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/03-02-04-Creating-agent_Access-control-security.jpg b/fern/docs/pages/agent-forge/screenshots/03-02-04-Creating-agent_Access-control-security.jpg new file mode 100644 index 0000000..9bcb206 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/03-02-04-Creating-agent_Access-control-security.jpg differ diff --git a/fern/docs/pages/agent-forge/screenshots/03-02-05-Creating-agent_Create-agent-page_topic-filter.png b/fern/docs/pages/agent-forge/screenshots/03-02-05-Creating-agent_Create-agent-page_topic-filter.png new file mode 100644 index 0000000..36bd2a8 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/03-02-05-Creating-agent_Create-agent-page_topic-filter.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/04-01-Knowledge-base-section.png b/fern/docs/pages/agent-forge/screenshots/04-01-Knowledge-base-section.png new file mode 100644 index 0000000..3f5e259 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/04-01-Knowledge-base-section.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/04-02-Knowledge-base-screen.png b/fern/docs/pages/agent-forge/screenshots/04-02-Knowledge-base-screen.png new file mode 100644 index 0000000..de24ebf Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/04-02-Knowledge-base-screen.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/04-03-01-Creating-kb_Create-New-KB.png b/fern/docs/pages/agent-forge/screenshots/04-03-01-Creating-kb_Create-New-KB.png new file mode 100644 index 0000000..c609c92 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/04-03-01-Creating-kb_Create-New-KB.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/04-03-02-Creating-kb_Upload-Files.png b/fern/docs/pages/agent-forge/screenshots/04-03-02-Creating-kb_Upload-Files.png new file mode 100644 index 0000000..a3ef7d5 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/04-03-02-Creating-kb_Upload-Files.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/04-03-03-Test_sim-matching_kb.png b/fern/docs/pages/agent-forge/screenshots/04-03-03-Test_sim-matching_kb.png new file mode 100644 index 0000000..e470f35 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/04-03-03-Test_sim-matching_kb.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/05-01-01-MCP-AgentBuilder-Select-Streaming-Mode.png b/fern/docs/pages/agent-forge/screenshots/05-01-01-MCP-AgentBuilder-Select-Streaming-Mode.png new file mode 100644 index 0000000..ae44432 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/05-01-01-MCP-AgentBuilder-Select-Streaming-Mode.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/05-01-02-MCP-AgentBuilder-Select-OrgLevelMCP_IfAvailable.png b/fern/docs/pages/agent-forge/screenshots/05-01-02-MCP-AgentBuilder-Select-OrgLevelMCP_IfAvailable.png new file mode 100644 index 0000000..e9e9555 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/05-01-02-MCP-AgentBuilder-Select-OrgLevelMCP_IfAvailable.png differ diff --git a/fern/docs/pages/agent-forge/screenshots/05-01-03-MCP-AgentBuilder-AddCustomMCP.png b/fern/docs/pages/agent-forge/screenshots/05-01-03-MCP-AgentBuilder-AddCustomMCP.png new file mode 100644 index 0000000..0cb3ef5 Binary files /dev/null and b/fern/docs/pages/agent-forge/screenshots/05-01-03-MCP-AgentBuilder-AddCustomMCP.png differ diff --git a/fern/docs/pages/welcome.mdx b/fern/docs/pages/welcome.mdx index e01a28f..52649ca 100644 --- a/fern/docs/pages/welcome.mdx +++ b/fern/docs/pages/welcome.mdx @@ -35,6 +35,19 @@ layout: custom {" "} + + Build and deploy AI agents without code. Create specialized assistants with knowledge bases, security controls, and custom configurations. + \ + \ + [**Get Started** →](/agent-forge/getting-started/overview) + + +{" "} +