Ollama Deep Research

Ollama Deep Research is a fully local web research assistant powered by any LLM hosted on Ollama.

Simply provide a research topic, and the assistant will intelligently generate web search queries, gather relevant results, and create comprehensive summaries. It then analyzes these summaries to identify knowledge gaps, generates follow-up queries to fill those gaps, and repeats this iterative process for a configurable number of research cycles.

The final output is a well-structured markdown report with cited sources from all research conducted.

✨ Features

• 🏠 Fully Local: No external API keys required - runs entirely on your machine with Ollama
• 🔄 Iterative Research: Automatically identifies knowledge gaps and conducts follow-up research loops
• 🔍 Smart Query Generation: LLM-powered search query optimization for better results
• 📊 Source Verification: Optional credibility scoring and assessment for research sources
• ⚙️ Highly Configurable: Adjust research depth, models, strategies, and output formats
• 📝 Multiple Output Formats: Support for Markdown, JSON, and HTML output
• 🎯 Research Strategies: Choose from broad overview, deep dive, or comparative analysis modes
• 🔗 Proper Citations: All findings include proper source citations with URLs
• ⚡ Parallel Processing: Asynchronous web searches for improved performance
• 🌐 Extensible Search: DuckDuckGo by default, with architecture for additional search APIs
• 🧠 Context-Aware: Maintains research context across multiple iterations
• 📈 Progress Tracking: Real-time logging of research steps and progress

🚀 Quickstart

Option 1: Docker (Recommended for Easy Setup)

1. Clone the project:

git clone https://github.com/Syed007Hassan/ollama_deep_research.git
cd ollama_deep_research

2. Copy environment configuration:

cp .env.example .env

3. Run with Docker Compose:

# Start both Ollama and the research assistant
docker-compose up -d

# Pull the model you want to use (after Ollama is running)
docker exec ollama-server ollama pull deepseek-r1:14b

# Check logs
docker-compose logs -f

4. Access the application:
   • LangGraph Studio: http://localhost:2024
   • Ollama API: http://localhost:11434

Option 2: Local Installation

1. Install Poetry (for dependency management):

pip install poetry

2. Clone and setup the project:

git clone https://github.com/Syed007Hassan/ollama_deep_research.git
cd ollama_deep_research

# Install dependencies with Poetry
poetry install

# Copy environment configuration
cp .env.example .env

Note: Poetry automatically creates and manages virtual environments, so you don't need to manually create one with python -m venv.

Selecting local model with Ollama

1. Download the Ollama app for Mac here.

2. Pull a local LLM from Ollama. As an example:

ollama pull deepseek-r1:14b

Selecting search tool

By default, it will use DuckDuckGo for web search, cause it does not require an API key.

⚙️ Configuration Options

You can configure the research assistant using environment variables in the .env file or through the LangGraph Studio UI. Below are all available configuration options:

Core Settings

Variable	Default	Description
LOCAL_LLM	deepseek-r1:14b	Name of the LLM model to use with Ollama
LLM_PROVIDER	ollama	LLM provider (currently only Ollama supported)
OLLAMA_BASE_URL	http://localhost:11434/	Base URL for Ollama API

Research Configuration

Variable	Default	Description
MAX_WEB_RESEARCH_LOOPS	3	Number of research iterations to perform
RESEARCH_STRATEGY	broad	Research approach: broad, deep, or comparative
MAX_SOURCES_PER_LOOP	3	Maximum number of sources to gather per research loop
ENABLE_SOURCE_VERIFICATION	false	Enable basic source credibility checking

Search Settings

Variable	Default	Description
SEARCH_API	duckduckgo	Web search API to use (currently only DuckDuckGo)
FETCH_FULL_PAGE	true	Include full page content in search results

Output Configuration

Variable	Default	Description
OUTPUT_FORMAT	markdown	Format for final output: markdown, json, or html
STRIP_THINKING_TOKENS	true	Remove <think> tokens from model responses

Example .env Configuration

# Core LLM Settings
LOCAL_LLM=qwen3:14b
OLLAMA_BASE_URL=http://localhost:11434/

# Research Configuration  
MAX_WEB_RESEARCH_LOOPS=5
RESEARCH_STRATEGY=deep
MAX_SOURCES_PER_LOOP=5
ENABLE_SOURCE_VERIFICATION=true

# Search Settings
SEARCH_API=duckduckgo
FETCH_FULL_PAGE=true

# Output Settings
OUTPUT_FORMAT=markdown
STRIP_THINKING_TOKENS=true

Configuration Priority

Keep in mind that configuration values are loaded in the following priority order:

1. Environment variables (highest priority)
2. LangGraph Studio UI configuration
3. Default values in the Configuration class (lowest priority)

Running with LangGraph Studio

Mac

1. Start Ollama service:

ollama serve

2. Launch LangGraph server with Poetry:

# Method 1: Using uv with Poetry
curl -LsSf https://astral.sh/uv/install.sh | sh
uvx --refresh --from "langgraph-cli[inmem]" --with-editable . --python 3.11 langgraph dev

# Method 2: Direct Poetry command
poetry run langgraph dev

Windows

1. Start Ollama service:

ollama serve

2. Launch LangGraph server with Poetry:

# Install LangGraph CLI in Poetry environment
poetry add --group dev "langgraph-cli[inmem]"

# Start the LangGraph server
poetry run langgraph dev

Using the LangGraph Studio UI

When you launch LangGraph server, you should see the following output and Studio will open in your browser:

Ready!

API: http://127.0.0.1:2024

Docs: http://127.0.0.1:2024/docs

LangGraph Studio Web UI: https://smith.langchain.com/studio/?baseUrl=http://127.0.0.1:2024

Open LangGraph Studio Web UI via the URL above. In the configuration tab, you can directly set various assistant configurations. Keep in mind that the priority order for configuration values is:

1. Environment variables (highest priority)
2. LangGraph UI configuration
3. Default values in the Configuration class (lowest priority)

How it works

Demo.mp4