An autonomous AI research agent that combines Ollama's web search with your local LLMs. Watch as your model reasons through complex queries, autonomously searches the web, and synthesizes intelligent answers with citations.

• Autonomous Agent - Local LLM decides when to search, fetch, and synthesize
• Clean Output - Shows final answers with a loading animation; optionally view full reasoning
• Smart Web Research - Autonomous web search and page fetching with context
• Fast & Efficient - Built with Rust for maximum performance
• Production Ready - Comprehensive error handling and logging
• Highly Configurable - Multiple models, output formats, and options

git clone https://github.com/guitaripod/weavex.git
cd weavex
cargo install --path .

cargo install weavex

You need an Ollama API key to use this tool. Get one at ollama.com/settings/keys.

export OLLAMA_API_KEY="your_api_key_here"

Or create a .env file:

echo "OLLAMA_API_KEY=your_api_key_here" > .env

Run autonomous research with your local Ollama models:

# Use default model (gpt-oss:20b) - opens result in browser by default
weavex agent "What are the top 3 Rust developments from 2025?"

# Disable browser preview, show in terminal
weavex agent --no-preview "query"

# Specify a different model
weavex agent --model qwen3:14b "research quantum computing trends"

# Show thinking steps and reasoning process for transparency
weavex agent --show-thinking "query"

# Disable model reasoning mode
weavex agent --disable-reasoning "query"

# Custom Ollama server
weavex agent --ollama-url http://192.168.1.100:11434 "query"

# Limit agent iterations
weavex agent --max-iterations 5 "query"

How it works:

1. Agent uses your local Ollama model for reasoning
2. Shows a loading animation (🧵 Weaving...) while working
3. Autonomously decides when to search the web or fetch URLs
4. Iterates until it has enough information
5. Opens the final result in your browser with markdown rendering (use --no-preview for terminal output)

Agent Output:

• 🧵 Weaving...: Loading animation while agent works
• 🌐 Browser Preview: Opens result in browser by default (use --no-preview for terminal output)

With --show-thinking flag:

• 🧠 Reasoning: Shows the model's thinking process
• 🔎 Searching: Web search operations
• 🌐 Fetching: URL fetch operations
• 💬 Response: Model's synthesized content

Requirements:

• Local Ollama server running (ollama serve)
• Model downloaded locally (ollama pull gpt-oss:20b)
• Ollama API key for web search access

Recommended Models:

• gpt-oss:20b - Best balance of speed and reasoning (default)
• qwen3:14b - Good tool-use capabilities
• qwen3:4b - Fastest, runs on laptops

# Opens results in browser by default
weavex "what is rust programming"

# Show results in terminal
weavex --no-preview "what is rust programming"

weavex --max-results 5 "best practices for async rust"

weavex --json "machine learning trends 2025"

weavex fetch https://example.com

# Pass API key via flag
weavex --api-key YOUR_KEY "query here"

# Verbose logging
weavex --verbose "debugging query"

  -k, --api-key <API_KEY>          Ollama API key (can also use OLLAMA_API_KEY env var)
  -m, --max-results <NUM>          Maximum number of search results to return
  -j, --json                       Output results as JSON
      --no-preview                 Disable browser preview (preview is enabled by default)
  -v, --verbose                    Enable verbose logging
      --timeout <SECONDS>          Request timeout in seconds [default: 30]
  -h, --help                       Print help
  -V, --version                    Print version

  fetch  Fetch and parse a specific URL
  agent  Run an AI agent with web search capabilities
  help   Print this message or the help of the given subcommand(s)

  -m, --model <MODEL>              Local Ollama model to use [default: gpt-oss:20b]
      --ollama-url <URL>           Local Ollama server URL [default: http://localhost:11434]
      --max-iterations <NUM>       Maximum agent iterations [default: 50]
      --show-thinking              Show agent thinking steps and reasoning process
      --disable-reasoning          Disable model reasoning (thinking mode)
      --no-preview                 Disable browser preview (preview is enabled by default)

# Opens result in browser by default
weavex agent "What are the latest benchmarks for Rust async runtimes?"

The agent will autonomously:

• Display a loading animation while working
• Search for relevant benchmark articles
• Fetch specific benchmark results
• Compare data from multiple sources
• Open the final result in your browser with markdown rendering

Show the reasoning steps and full transparency:

weavex agent --show-thinking "What are the latest benchmarks for Rust async runtimes?"

Disable browser preview to see output in terminal:

weavex agent --no-preview "What are the latest benchmarks for Rust async runtimes?"

Disable reasoning mode for faster responses:

weavex agent --disable-reasoning "What are the latest benchmarks for Rust async runtimes?"

weavex "latest rust async runtime benchmarks"

weavex --max-results 10 "tokio vs async-std performance"

weavex fetch https://blog.rust-lang.org/

weavex --json "rust web frameworks" | jq '.results[0].url'

cargo build

cargo test

cargo build --release

src/
├── main.rs        - Application entry point and orchestration
├── agent.rs       - AI agent loop with tool execution
├── cli.rs         - CLI argument parsing with clap
├── client.rs      - Ollama web search API client
├── config.rs      - Configuration management
├── error.rs       - Custom error types with thiserror
├── formatter.rs   - Output formatting (human & JSON)
└── ollama_local.rs - Local Ollama chat API client

Contributions are welcome! Please feel free to submit a Pull Request.

Built with:

• clap - CLI argument parsing
• reqwest - HTTP client
• tokio - Async runtime
• serde - Serialization framework

Powered by Ollama's Web Search API.