For the complete documentation index, see llms.txt. Markdown variants are available by appending .md to any URL or sending an Accept: text/markdown header. An agent skill is available at /.well-known/agent-skills/site-skill.md.
319
Sponsor

Deep Search

PreviousNext

A multi-agent deep search system that researches questions by clarifying intent, planning queries, evaluating results, and synthesizing answers.

EveFlueMastra

Output will stream here when you run the agent.

Summary

The Deep Search Agent is a multi-agent system that researches questions through an iterative process. It clarifies user intent, plans search queries, evaluates results for completeness, and synthesizes a comprehensive answer. The system uses 4 specialized agents working together in a workflow to produce well-researched, cited answers.

Installation

$ pnpm dlx shadcn@latest add @agentcn/mastra/deep-search

Composition

├── config.ts                          # Main agent config with workflow
├── instructions.md                    # Main agent instructions
├── subagents/
│   ├── intent-clarifier/
│   │   ├── config.ts                  # Intent clarifier config
│   │   └── instructions.md            # Generates clarifying questions
│   ├── research-planner/
│   │   ├── config.ts                  # Research planner config
│   │   └── instructions.md            # Writes search queries
│   ├── search-result-evaluator/
│   │   ├── config.ts                  # Search evaluator config
│   │   └── instructions.md            # Evaluates result quality
│   └── answerer/
│       ├── config.ts                  # Answerer config
│       └── instructions.md            # Synthesizes final answer
├── skills/
│   └── research/
│       └── SKILL.md                   # Research procedure
└── workflows/
    └── deep-search.ts                 # Orchestrates all agents

How It Works

  1. Clarify Intent - The intent-clarifier agent asks 3 clarifying questions to understand the user's needs
  2. Plan Queries - The research-planner agent generates 3-5 search queries based on the clarified intent
  3. Search - The system executes searches using Bright Data SERP API
  4. Evaluate - The search-result-evaluator agent checks if results are sufficient
  5. Iterate - If results are insufficient, the system loops back to step 2 (up to 4 iterations)
  6. Answer - The answerer agent synthesizes a comprehensive, cited answer

Environment Variables

BRIGHTDATA_API_KEY=       # Required: Bright Data API key
BRIGHTDATA_SERP_ZONE=     # Optional: Bright Data SERP zone (defaults to "serp_api")

Customization

  • Search provider. Replace the performSearch function in workflows/deep-search.ts with another provider like Exa, SerpAPI, or Google Custom Search.
  • Swap models. Edit the model field in each subagent's config.ts.
  • Adjust iterations. Modify the loop limit in workflows/deep-search.ts (currently 4).
  • Add more agents. Create new subagents under subagents/ and wire them into the workflow.