logo
Sign In

OWUIAgenticToolingSuite/Exa Agentic Search

tool

Elevate your AI's research capabilities with intelligent, multi-modal web search.

sdjfhsud
@sdjfhsud
·
6 months ago
312
6
4.5K
Updated 6 months ago
Created Jul 21, 2025
exa_router_search.py
Python · 1123 lines
"""
Title: Exa Agentic Search Tool
Description: An intelligent agentic research tool with full control over Exa search parameters.
author: ShaoRou459
author_url: https://github.com/ShaoRou459
Version: 2.0.1
Requirements: exa_py, open_webui
"""

from __future__ import annotations

import os

Exa Agentic Search Tool

Intelligent agentic research with AI-controlled search parameters


📚 Full Documentation: GitHub Repository

What This Does

The Exa Agentic Search is a complete rewrite of the search system with an agentic architecture. Instead of rigid search modes, the AI agent now has full control over search parameters—dynamically adapting search type, categories, date filters, domain filters, and result counts based on your query complexity.

🔗 Works Best With

Auto Tool Selector (Function) - Install this companion function for automatic tool activation and enhanced capabilities.

Key Features

  • Agent-Controlled Parameters - AI decides search type, category, date ranges, and domains.
  • Iterative Research - Multi-pass search with intelligent stopping criteria.
  • Smart Query Generation - Agent generates optimized queries based on context.
  • Real-Time Status - Live progress updates keep you informed throughout the agent's thought and search process.
  • Enhanced Debug System - Clean, focused logging with session summaries.
  • Unified API - Single efficient API call with search_and_contents.

What's New in v2.0.0

  • Agentic Architecture: The AI controls all search parameters, not rigid modes.
  • Intelligent Stopping: Agent decides when enough information is gathered.
  • Real-Time UI Status: Watch the agent iterate through research steps live in your OpenWebUI chat.
  • Standardized Output: Consistent format (EXTRACTED_INFO → EVALUATION → DECISION → SEARCH_CONFIG).
  • Updated Model Support: Optimized for the latest advanced reasoning models like GPT-5.4 Thinking, Gemini 3.1 Pro, and DeepSeek.
  • Simplified Config: Just 4 core settings for easier management instead of 6+.

Prerequisites

Required: Exa API Key

Get your free API key at exa.ai and add it to the exa_api_key setting.

Recommended: Auto Tool Selector

This tool works best with the Auto Tool Selector Function for automatic, seamless activation without manual toggling.

Installation

  1. Install Dependencies: Run pip install exa_py inside your OpenWebUI Docker container.
  2. Install Tool: Go to Workspace → Tools → New Tool and paste the exa_agentic_search.py code.
  3. Set Tool ID to: exa_agentic_search (important!)
  4. Add your Exa API key in tool settings.
  5. Install the companion function: Auto Tool Selector.

How the Agent Works

The agent intelligently adapts to your query:

Simple Queries

"What is the capital of France?" → Quick search, finds answer, stops after 1 iteration (~10 seconds)

Research Queries

"Compare the latest developments in quantum computing from different research labs" → Multi-iteration search, comprehensive synthesis (~30-60 seconds)

The Agent Controls

ParameterOptionsExample
search_typeauto, neural, fast, deepneural for semantic search
categorynews, research paper, github, company, tweet, pdf, etc.research paper for academic queries
num_results1-25More results for research, fewer for simple facts
date_filtersstart/end datesRecent news vs historical data
domain_filtersinclude/exclude domainsFocus on specific sources

Key Settings

SettingPurposeRecommendation
exa_api_keyRequired: Your Exa API keyGet from exa.ai
agent_modelControls search strategy & evaluationGPT-5.4 Thinking, Gemini 3.1 Pro, or local DeepSeek
max_iterationsMaximum search passes before stopping3 (default)
debug_enabledDetailed operation logsfalse (enable for troubleshooting)
show_sourcesDisplay source citations in UIfalse (optional)

Debug Output

When debug is enabled, you'll see clean, focused output in your Docker logs (docker logs open-webui -f):

════════════════════════════════════════════════════════════ 🔍 AGENTIC SEARCH [1234] ════════════════════════════════════════════════════════════ Query: quantum computing developments ▶ ITERATION 1/3 ├─ Config: │ Query: quantum computing 2026 breakthroughs... │ Type: auto | Cat: news | Num: 10 ├─ Results: 8 docs │ • nature.com - Breakthrough in quantum... └─ Decision: ✓ STOP 📊 SESSION SUMMARY Stats: 2.5s total | 2 LLM calls | 8 sources ════════════════════════════════════════════════════════════

Usage Tips

  • Let the agent decide: No need to specify search modes—it adapts automatically.
  • Complex queries: The agent will iterate and refine searches as needed.
  • Real-Time Feedback: Watch the UI status to see what search categories and keywords the agent is evaluating.
  • Debug issues: Enable debug_enabled and check Docker logs.
  • Performance: Simple queries stop early, complex ones get thorough research.

Migrating from Exa Router Search

If you're upgrading from the previous version:

  1. Update the tool code with the new exa_agentic_search.py.
  2. Ensure the Tool ID is set to exa_agentic_search (update your Auto Tool Selector settings if you changed it).
  3. Settings are simpler now—agent_model replaces the multiple model-specific settings.

Elevate your AI's research capabilities with intelligent, agentic web search.

2