Tools Overview

Extend your agents with powerful tools and capabilities

What Are Tools in Junis?

Tools are specialized functions that agents can call to perform specific actions or access external data. They extend your agent’s capabilities beyond text generation, enabling real-world interactions.

Think of tools as skills your agents can learn:

• 🔍 Search tools: Find information from web, databases, or documents
• 📄 Document tools: Process PDF, Excel, Word, PowerPoint files
• 🔌 External connectors: MCP, Composio OAuth (200+ apps), GitHub, Firecrawl
• 🧠 RAG tools: Search and create knowledge base documents with AI-powered retrieval
• 💾 Memory tools: Read and update persistent user memory across conversations
• ✨ Soul tools: Read and update organization AI personality

Agent Intelligence: Agents automatically decide when and how to use tools based on user requests. You don’t need to manually trigger them.

---

Tool Types

Junis supports four types of tools, each designed for specific use cases:

Regular Tools

Python Functions

Custom async functions you build for your specific needs.

Examples: Search databases, call APIs, process data

Claude Skills

Official Anthropic Tools

Prebuilt document processors powered by Claude.

Examples: PDF extraction, Excel analysis, PPTX creation

External Connectors

Third-Party Integrations

Connect to external systems and services.

Examples: Salesforce, DataLab, GitHub, Firecrawl (via MCP)

RAG Tools

Document Retrieval

Auto-generated search tools for uploaded documents.

Examples: Company policies, employee handbooks, product specs

---

How Agents Use Tools

🤖 Automatic Tool Selection

Agents use tools intelligently based on:

1. User Request: What the user is asking for
2. Tool Descriptions: What each tool can do (from docstrings)
3. Conversation Context: Previous messages and results

Example Workflow:

User: "Find the latest news about artificial intelligence"
  ↓
Agent thinks:
  - User wants to search the web for current information
  - I have a tool called "google_search"
  - Its description says it searches the web for real-time information
  - I should use this tool with query="latest artificial intelligence news"
  ↓
Agent calls: google_search("latest artificial intelligence news")
  ↓
Tool returns: {
  "results": [
    {
      "title": "AI Breakthrough in 2026",
      "url": "https://example.com/ai-news",
      "snippet": "Recent developments in AI technology..."
    }
  ]
}
  ↓
Agent responds: "I found recent AI news. Here are the latest developments: AI Breakthrough in 2026..."

No Manual Intervention: Once tools are configured, agents use them automatically without user or admin prompting.

📊 Tool Execution Flow

---

Available Built-in Tools

Junis provides a comprehensive set of prebuilt tools ready to use:

Document Processing (Claude Skills)

PDF

claude_pdf_processor

• Extract text from PDFs
• Extract tables and data
• Create new PDFs
• Merge multiple PDFs
• Convert to images

Use Case: “Extract all customer information from this invoice PDF”

Excel

claude_xlsx_processor

• Analyze spreadsheets
• Extract specific columns/rows
• Generate charts
• Perform calculations
• Export filtered data

Use Case: “Calculate total revenue from this sales spreadsheet”

PowerPoint

claude_pptx_creator

• Create presentations
• Add slides with content
• Insert images and charts
• Apply templates
• Export to PDF

Use Case: “Create a 5-slide presentation about our Q4 results”

Word

claude_docx_processor

• Extract text from Word docs
• Create new documents
• Track changes
• Manage comments
• Convert to other formats

Use Case: “Extract all action items from this meeting notes document”

Search Tools

Web Search

google_search

Search the web for real-time information using Google.

Parameters:

• query (str): Search query

Returns: Search results with titles, URLs, and snippets

Use Case: “Find the latest AI news” or “Search for Python tutorials”

Built-in Web Search (LLM Feature)

NEW: Agents can now perform web searches automatically during response generation without explicitly calling a tool.

Enable Web Search in Admin > Agents > Edit Agent:

Agent Configuration:
  Enable Web Search: ✅ ON

How It Works:

• When enabled, the LLM can automatically search the web during response generation
• No explicit tool call required - the model decides when to search
• Uses LiteLLM’s web_search_options with search_context_size: "medium"

Difference from google_search Tool:

Feature	Built-in Web Search	google_search Tool
Activation	Automatic (LLM decides)	Explicit tool call
Configuration	Toggle in Agent settings	Assign tool to agent
Best For	General knowledge queries	Targeted searches

When to Use: Enable for agents that need to answer questions about current events, recent news, or real-time information without explicit search requests.

Memory & Soul Tools (Auto-Injected)

Automatic: These tools are automatically available to all orchestrator agents. No configuration needed.

User Memory

update_user_memory(action, content)

Read or update the user’s persistent profile (name, role, preferences).

Actions:

• read: View the user’s current profile
• update: Save updated profile content

Use Case: Agent learns user’s name, role, or communication preferences during conversation

Eternal Memory

save_to_eternal_memory(action, content)

Store permanent facts and standing instructions that persist across all conversations.

Actions:

• read: View stored eternal memories
• add: Append a new memory entry
• update: Replace all eternal memory content

Use Case: User says “remember that I always want bullet points” or “don’t forget my deadline is Friday”

Soul

update_soul(action, content)

Read or update the organization’s AI thinking philosophy (Admin/Owner only).

Use Case: Admin asks “update our AI to prioritize data-driven decisions”

Identity

update_identity(action, content)

Read or update the organization’s AI communication style (Admin/Owner only).

Use Case: Admin asks “make our AI more casual and friendly”

Learn more about these systems: Memory System | Soul & Identity

RAG Document Upload (Agent Tool — Requires Configuration)

upload_rag_document(datastore_name, file_name, content, description?)

Agents can generate and store their own knowledge base documents as Markdown files. This is unique to Junis — your AI agents can research, synthesize, and save information for future retrieval.

Configuration Required: Unlike memory and soul tools (which are auto-injected), upload_rag_document must be explicitly assigned to an agent via Admin > Agents > Edit Agent > Tools.

How it works:

1. Agent generates content (e.g., research summary, analysis report)
2. Agent calls upload_rag_document with the content as Markdown
3. A .md file is created, uploaded to a RAG DataStore, and indexed
4. Future conversations can search and retrieve this knowledge

Use Case: “Research the latest AI trends and save a summary to our knowledge base”

Agent:
1. Searches the web for AI trends
2. Synthesizes findings into a structured document
3. Calls upload_rag_document(
     datastore_name="Research Library",
     file_name="ai-trends-q1-2026",
     content="# AI Trends Q1 2026\n\n## Key Developments..."
   )
4. Document is indexed and searchable by all agents

DataStore Auto-Creation: If the specified DataStore doesn’t exist, the tool creates it automatically. The file extension .md is auto-appended if not provided.

Payment Tools (AP2/x402 Protocol)

Premium Feature: Payment tools enable USDC-based agent usage fees with 80/20 revenue split (Agent Owner 80%, Platform 20%).

require_payment

require_payment(agent_id, network?)

Request payment before agent execution.

When to Use:

• Premium agents that charge per use
• Computationally expensive operations
• Third-party API costs

Returns:

• Payment requirements (amount, wallet, network)
• EIP-3009 nonce for signature
• 80/20 split details

verify_payment

verify_payment(nonce, payment_type)

Verify EIP-3009 payment signatures.

Payment Types:

• agent_owner: 80% to agent owner
• platform_fee: 20% to platform

Auto-Settlement: When both payments verified, automatically settles on-chain.

---

When to Use Tools

✅ Use Tools When You Need To:

• Access External Data: Databases, APIs, file storage
• Perform Complex Operations: File processing, calculations, analysis
• Integrate Third-Party Services: Salesforce, GitHub, Firecrawl
• Retrieve Uploaded Documents: RAG search through company docs

❌ Don’t Build Tools For:

• Simple Text Generation: Agents already excel at this
• One-Time Tasks: Better to do manually
• Highly Variable Logic: Better in agent instructions
• Duplicate Functionality: Use existing tools or MCP platforms

---

Getting Started with Tools

Step 1: Choose Your Approach

Use Built-in Tools

Fastest: Junis provides 15+ prebuilt tools.

1. Go to Admin > Agents
2. Edit your agent
3. Select tools from the Tools section
4. Save

Done! Your agent can now use those tools.

Connect MCP Platforms

Recommended: Use Model Context Protocol for external services.

1. Go to Team > MCP
2. Connect platforms (GitHub, Firecrawl, etc.)
3. Assign to agents

See MCP Integration Guide for details.

Build Custom Tools

For Advanced Use Cases: Junis supports custom tool development for specialized business logic that isn’t covered by built-in tools.

Process:

1. Contact Junis Support at [e-mail]
2. Describe your use case and requirements
3. Our team will develop and register the custom tool
4. You assign it to your agents via Admin UI

Note: Most users don’t need custom tools - the built-in library covers common use cases.

Upload Documents (RAG)

For Knowledge Bases: Upload PDFs, docs, etc.

1. Go to Team > RAG Knowledge
2. Create or select DataStore
3. Upload files
4. Assign RAG tool to agent

See RAG Guide for details.

Step 2: Configure Agent Tools

Admin Dashboard → Agents → Edit Agent

Agent Configuration:
  Name: Customer Support Agent
  Type: LLM_AGENT
  Tools:
    Regular Tools:
      - claude_pdf_processor
      - google_search
    RAG Tools:
      - rag_company-policies_1234567890
  MCP Platforms:
    - GitHub (for repository management)
    - Firecrawl (for web scraping)

Step 3: Test with Chat

You: "Search for recent AI trends and create a GitHub issue to track them"

Agent:
1. Uses google_search("recent AI trends 2026")
2. Formats the results
3. Uses GitHub MCP to create issue

Response: "I found the latest AI trends and created a tracking issue in your repository."

---

Tool Best Practices

✅ DO:

• Choose specific tools for specific tasks
• Write clear tool descriptions (docstrings)
• Test tools individually before assigning to agents
• Monitor tool usage in logs
• Handle errors gracefully
• Use RAG for frequently accessed documents

❌ DON’T:

• Add too many tools (max 10 per agent recommended)
• Use tools with unclear purposes
• Forget to update tool descriptions
• Ignore tool errors in production
• Hard-code sensitive data in tools

---

Real-World Examples

1. Customer Support Agent

Tools:

• claude_pdf_processor: Extract info from uploaded docs
• GitHub MCP: Create support tickets as issues
• rag_faq_documents: Search company FAQ

User: “A customer asked about our return policy”

Agent Flow:

1. Search rag_faq_documents for “return policy”
2. Extract relevant sections
3. Format a friendly response
4. Optionally create GitHub issue for tracking

2. Data Analysis Agent

Tools:

• claude_xlsx_processor: Analyze Excel files
• google_search: Find market trends and research
• claude_pptx_creator: Generate presentation

User: “Analyze Q4 sales and create a presentation”

Agent Flow:

1. Use claude_xlsx_processor to analyze sales spreadsheet
2. Use google_search for market research and context
3. Use claude_pptx_creator to generate slides

3. Premium Agent with Payments

Tools:

• require_payment: Gate agent access with USDC payment
• claude_pdf_processor: Document processing
• google_search: Research capabilities

User: “Run premium analysis workflow”

Agent Flow:

1. Call require_payment(agent_id="premium_agent")
2. Wait for user payment (MetaMask)
3. Verify payment signatures
4. Proceed with premium analysis using available tools

---

Tool Categories Reference

Category	Tools	Primary Use
Document Processing	claude_pdf_processor, claude_xlsx_processor, claude_pptx_creator, claude_docx_processor	File analysis, creation, extraction
Web Search	google_search, built-in web search	Real-time web information retrieval
Memory	update_user_memory, save_to_eternal_memory	Persistent user memory management
Soul & Identity	update_soul, update_identity	Organization AI personality
RAG Search	rag_ (auto-generated)	Document knowledge retrieval
RAG Upload	upload_rag_document	Agent-generated knowledge creation
Payment	require_payment, verify_payment	Premium agent gating
External	MCP platforms, Composio OAuth	Third-party service integration

---

Performance Considerations

⚡ Tool Execution Speed

Tool Type	Typical Latency	Notes
Database Query	100-500ms	Depends on query complexity
External API	1-5 seconds	Network dependent
Claude Skills	2-10 seconds	File processing time
RAG Search	1-3 seconds	Vertex AI Search latency
Payment	5-30 seconds	User signature + on-chain settlement

Optimization Tip: Group multiple simple queries into batch tools rather than making agents call tools repeatedly.

---

Troubleshooting

Agent not using available tools

Symptom: Tool is configured but agent doesn’t call it

Possible Causes:

• Tool description (docstring) unclear
• Agent instruction doesn’t mention tool usage
• Tool name not intuitive

Solution:

1. Review tool docstring for clarity
2. Add tool usage examples to agent instruction
3. Test with explicit request: “Use [tool_name] to…”

Tool execution fails silently

Symptom: Tool called but no result returned

Possible Causes:

• Tool has unhandled exception
• Missing required parameters
• External service down

Solution:

1. Check agent session logs for errors
2. Test tool directly via API
3. Verify all environment variables set
4. Check external service status

RAG tool returns no results

Symptom: RAG search completes but finds nothing

Possible Causes:

• Documents not indexed yet (still processing)
• Search query too specific
• DataStore not created properly

Solution:

1. Go to Team > RAG Knowledge and check indexing status
2. Wait for “COMPLETED” status (can take 5-60 minutes)
3. Try broader search terms
4. Re-upload documents if status is “FAILED”

Payment tool timeout

Symptom: require_payment call takes too long

Possible Causes:

• Network congestion (Base blockchain)
• CDP Facilitator API slow
• User hasn’t signed MetaMask prompt

Solution:

• Payment tools have 30-second timeout
• Check Base network status: https://status.base.org
• Ensure user completes MetaMask signature promptly
• For testnet, use Base Sepolia (faster, free)

---

What’s Next?

Memory System

Persistent AI memory across conversations

Soul & Identity

Define your AI team’s personality

RAG Setup

Upload documents and create knowledge bases

MCP Platforms

Connect external services via MCP and Composio

---

Additional Resources

• Claude Skills: https://docs.anthropic.com/en/docs/build-with-claude/tool-use - Official Anthropic documentation on tool use

Need Help? Contact us at [e-mail] for assistance with tool development and custom implementations.