AI Search API for CrewAI - Contextual Prompts Guide

A Python integration for CrewAI that connects your agents to the AI Search API.
It enables contextual prompts, multi-message context handling, and intelligent answers with source citations.

👉 Get your free API key from the AI Search API dashboard.

Features

• 🔍 Prompt + Context Search: Send a query plus structured context for better results
• 💬 Multi-Message Context: Pass multiple user messages for conversational awareness
• 📚 Sources Included: Responses come with citations when available
• ⚡ CrewAI Integration: Works with Agent, Task, Crew out of the box
• 🖥️ Works with Local LLMs: Use Ollama for reasoning + the search API for live info
• 🛡️ Robust Error Handling: Clear exceptions when context roles or models are invalid

Installation

pip install crewai-aisearchapi

Quick Start (with Ollama + CrewAI)

from crewai import Agent, Task, Crew, Process, LLM
from crewai_aisearchapi import AISearchTool

llm = LLM(
    model="ollama/llama3.2:3b",          # use a local model
    base_url="http://localhost:11434",   # Ollama default
    temperature=0.2,
)

tool = AISearchTool(api_key="your-api-key")

agent = Agent(
    role="Researcher",
    goal="Answer questions with context and sources.",
    backstory="Careful and concise.",
    tools=[tool],
    llm=llm,
    verbose=True,
)

task = Task(
    description="Answer: '{question}'. Keep it short.",
    expected_output="2–4 sentences.",
    agent=agent,
    markdown=True,
)

crew = Crew(agents=[agent], tasks=[task], process=Process.sequential, verbose=True)

if __name__ == "__main__":
    print(crew.kickoff(inputs={"question": "What is RLHF in AI?"}))

Contextual Prompts

The AI Search API is designed to handle context-rich prompts.
All context messages must use "role": "user".

result = tool.run({
    "prompt": "Explain how RLHF improves AI safety.",
    "context": [
        {"role": "user", "content": "Keep it simple, I'm new to ML."},
        {"role": "user", "content": "Add one practical example."}
    ],
    "response_type": "markdown"
})

This produces an answer that considers both your question and the context messages.

Configuration Options

from crewai_aisearchapi import AISearchTool, AISearchToolConfig

config = AISearchToolConfig(
    default_response_type="markdown",
    include_sources=True,
    timeout=30,
    verbose=True
)

tool = AISearchTool(api_key="your-api-key", config=config)

Handling Responses

The tool returns a formatted string containing:

• Answer (the AI response)
• Sources (when available)
• Response Time

Example:

Reinforcement Learning with Human Feedback (RLHF) helps align AI models with human intent...

**Sources:**
- [1] https://example.com/rlhf-overview
- [2] https://research.example.org/rlhf

*Response time: 120ms*

Environment Variables

export AISEARCH_API_KEY="your-api-key"

In Python:

import os
from crewai_aisearchapi import AISearchTool

tool = AISearchTool(api_key=os.getenv("AISEARCH_API_KEY"))

Troubleshooting

• model not found → Ensure you pulled the Ollama model:
  ollama pull llama3.2:3b
• Each context message must have role "user" → Convert system → user in your context.
• API key errors → Check AISEARCH_API_KEY is set correctly in your environment.

License

MIT License - see the LICENSE file.

Support

• Dashboard & API Key: AI Search API Dashboard
• Docs: docs.aisearchapi.io
• Homepage: aisearchapi.io
• Issues: GitHub Issues