Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/danny-avila/LibreChat/llms.txt

Use this file to discover all available pages before exploring further.

​
Overview

Agents in LibreChat are powerful AI assistants that can use tools, execute code, search the web, and perform complex multi-step tasks. They combine language models with extensible capabilities through the agent builder interface.

​
What Are Agents?

Agents extend basic chat functionality by:
  • Using Tools: Execute code, search the web, generate images, and access custom APIs
  • Multi-Step Reasoning: Break down complex tasks into sequential steps
  • File Search: Retrieve information from uploaded documents
  • Custom Actions: Connect to external APIs via OpenAPI specifications
  • Agent Handoffs: Chain multiple specialized agents together

​
Creating an Agent

1

Open the Agent Builder

Click the Agent button in the side panel or select an existing agent to modify.
2

Configure Basic Settings

Set the agent’s name, description, and custom instructions that guide its behavior.
# Example agent configuration
name: "Research Assistant"
description: "Helps research topics and summarize findings"
instructions: "You are a research assistant. Search the web for accurate information and provide detailed summaries with citations."
3

Select Model and Provider

Choose the AI model that powers your agent. Different models offer different capabilities:
  • GPT-4: Best for complex reasoning
  • Claude: Excellent for analysis and writing
  • Gemini: Strong multimodal capabilities
4

Enable Capabilities

Toggle the tools and capabilities your agent needs:
  • Code Interpreter: Execute Python code
  • Web Search: Search and retrieve web content
  • File Search: Query uploaded documents
  • Image Generation: Create images with DALL-E, Flux, or Stable Diffusion
  • Actions: Connect to external APIs
5

Configure Advanced Settings (Optional)

Fine-tune behavior with advanced options:
  • Recursion Limit: Maximum steps for multi-step tasks (default: 25)
  • Model Parameters: Temperature, top_p, max tokens
  • Agent Handoffs: Chain multiple agents together
  • Sequential Outputs: Show or hide intermediate reasoning steps

​
Agent Capabilities

​
Code Interpreter

Execute Python code in a secure sandboxed environment. 
# librechat.yaml
endpoints:
  agents:
    capabilities: ["execute_code"]
API key required for code execution:
# .env
LIBRECHAT_CODE_API_KEY=your-key
Search the internet and retrieve relevant information. 
# librechat.yaml
webSearch:
  serperApiKey: '${SERPER_API_KEY}'     # Search provider
  firecrawlApiKey: '${FIRECRAWL_API_KEY}' # Content scraper
  jinaApiKey: '${JINA_API_KEY}'         # Result reranker
Query uploaded documents using vector search and retrieval.
File search uses RAG (Retrieval Augmented Generation) to find relevant information in your documents. Configure citation settings in librechat.yaml:
agents:
  maxCitations: 30
  maxCitationsPerFile: 7
  minRelevanceScore: 0.45  # 0.0 to 1.0

​
Actions (Custom APIs)

Connect agents to external services via OpenAPI specifications.
1

Prepare OpenAPI Spec

Create or import an OpenAPI 3.x specification for your API.
2

Configure Domain Allowlist

For security, explicitly allow API domains:
# librechat.yaml
actions:
  allowedDomains:
    - 'api.example.com'
    - 'https://internal.api:8443'  # With protocol/port
3

Add Action to Agent

In the agent builder, paste the OpenAPI spec URL or upload the specification file.
By default, SSRF targets (localhost, private IPs, .internal/.local TLDs) are blocked. You must explicitly add them to allowedDomains to enable access.

​
Advanced Features

​
Recursion Limit

Control how many steps an agent can take: 
# librechat.yaml
endpoints:
  agents:
    recursionLimit: 50        # Default for all agents
    maxRecursionLimit: 100    # Maximum allowed

​
Agent Handoffs

Chain multiple specialized agents together for complex workflows: 
// Example: Research Agent → Writer Agent → Editor Agent
edges: [
  { source: "research-agent-id", target: "writer-agent-id" },
  { source: "writer-agent-id", target: "editor-agent-id" }
]

​
Model Parameters

Fine-tune model behavior: 
{
  "temperature": 0.7,
  "top_p": 0.9,
  "max_tokens": 4096,
  "frequency_penalty": 0.0,
  "presence_penalty": 0.0
}

​
Sharing Agents

Share agents with specific users or groups:
  1. Click Share in the agent footer
  2. Select users, groups, or roles
  3. Click Share

​
Agent Categories

Organize agents by category for easier discovery:
  • Productivity: Task management, scheduling, email
  • Research: Information gathering, analysis
  • Creative: Writing, design, content creation
  • Technical: Coding, debugging, DevOps
  • Custom: Your own categories

​
Configuration Reference

# librechat.yaml
interface:
  agents:
    use: true      # Allow users to use agents
    create: true   # Allow users to create agents
    share: true    # Allow private sharing
    public: false  # Allow public sharing

endpoints:
  agents:
    disableBuilder: false
    recursionLimit: 25
    maxRecursionLimit: 100
    maxCitations: 30
    maxCitationsPerFile: 7
    minRelevanceScore: 0.45
    capabilities:
      - execute_code
      - file_search
      - web_search
      - actions
      - tools

​
Best Practices

  • Clear Instructions: Write specific, actionable instructions for your agent
  • Right Tools: Only enable capabilities the agent needs
  • Test Iteratively: Start simple and add complexity gradually
  • Monitor Tokens: Agents with multiple steps use more tokens
  • Set Limits: Use recursion limits to prevent runaway processes

Build docs developers (and LLMs) love

Get started for freeTalk to us