> For the complete documentation index, see [llms.txt](https://docs.graphlit.dev/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.graphlit.dev/examples/deep-research.md).

# Deep Research

Learn to build autonomous AI research agents that perform multi-hop web research, entity extraction, and knowledge synthesis.

## Choose Your Framework

Build the same deep research agent with your preferred orchestration framework:

### [Mastra (TypeScript)](/examples/deep-research/mastra.md)

**Best for**: TypeScript developers, modern web apps

⏱️ **Time**: 30-40 minutes\
🎯 **Level**: Advanced

**What you'll build**:

* Entity-driven research using knowledge graphs
* Pre-ingestion filtering with native reranking
* Autonomous convergence detection
* Multi-source synthesis with citations

### [Agno (Python)](/examples/deep-research/agno.md)

**Best for**: Python developers, high-performance systems

⏱️ **Time**: 30-40 minutes\
🎯 **Level**: Advanced

**Why Agno**:

* 5000x faster than LangGraph
* 50x less memory usage
* Simpler code (just Python functions!)
* Same algorithm, cleaner implementation

***

## What All Tutorials Cover

Every tutorial teaches the same 5-phase research algorithm, implemented in different frameworks:

### Phase 1: Seed Acquisition

Start from a URL or search query to establish initial knowledge base

### Phase 2: Entity-Driven Discovery

Extract entities from your knowledge graph (Person, Organization, Category)

### Phase 3: Intelligent Expansion

Search web for each entity, **filter before ingesting** (key innovation!)

### Phase 4: Convergence Detection

Automatically detect when research has converged (novelty scoring)

### Phase 5: Multi-Source Synthesis

Generate comprehensive reports from 100+ sources using summary-based RAG

***

## Key Innovations

**1. Pre-Ingestion Filtering**

* Analyze 50 sources, ingest only top 8
* Uses Graphlit's native reranker
* Significantly faster with better quality

**2. Diminishing Returns Detection**

* Agent knows when to stop researching
* Based on novelty scoring (% new sources in top 10)
* No manual intervention needed

**3. Summary-Based RAG**

* Scales beyond traditional RAG (10-20 docs → 100+)
* Operates on optimized summaries
* Fast and accurate

***

## What Graphlit Provides

All frameworks use the same Graphlit SDK:

✅ Automatic entity extraction (during ingestion)\
✅ Knowledge graph (Schema.org/JSON-LD)\
✅ Native reranker (enables pre-filtering)\
✅ Exa search (built-in, no API key needed)\
✅ Summary-based RAG (scalable synthesis)\
✅ Multi-source citations

**Time saved**: 12-14 weeks of infrastructure development

***

## Coming Soon

More framework tutorials:

* **LangGraph (Python)** - Graph-based state machines
* **Vercel Workflow (TypeScript)** - Deterministic, durable orchestration

***

## Choose Your Tutorial

**TypeScript developer?** → [Start with Mastra](/examples/deep-research/mastra.md)

**Python developer?** → [Start with Agno](/examples/deep-research/agno.md)