TIGER-AI-Lab Releases OpenResearcher, an API-Free Open-Source Recipe for Training Deep Research Agents

TIGER-AI-Lab has published OpenResearcher, a fully open-source pipeline that generates roughly 96,000 high-quality deep research trajectories and trains agentic language models on long-horizon web research tasks without relying on any external search APIs. The project bundles a dataset, model checkpoints, a reproducible training and evaluation recipe, and a self-contained offline search environment, positioning it as one of the more complete open blueprints for building autonomous research agents.