SearchOS

What is SearchOS?

SearchOS is a joint initiative launched by the Hillhouse School of Artificial Intelligence at Renmin University of China and Ant Group‌Open Source Plusintelligent bodySearch Collaboration Framework‌, it abstracts open-domain information retrieval into a dynamic relational model, addressing system-level issues in long-range search tasks—such as agent amnesia, redundant queries, and source decoupling—and provides a system-level multi-agent collaboration infrastructure for search agents.

Key Features of SearchOS

• ‌Building Relational Search Models‌: Automatically converts natural language queries into a dynamic relational model with table structures, primary keys, and foreign keys, while automatically completing attributes as entities are discovered.
• ‌Multi-Agent Collaborative Search‌: The Orchestrator handles centralized scheduling; the Explore Agent discovers entities; the Search Agent executes parallel tasks; and the Writer Agent generates structured reports.
• ‌Context Management for Search‌: Maintains four types of shared states—frontier tasks, evidence graphs, coverage maps, and failure memories—to enable progress synchronization across agents.
• ‌Search Tool Middle Tier‌: Complete state injection, loop detection, and structured evidence extraction; intercept invalid operations.
• ‌Hierarchical Skills Database‌: Includes approximately 280 predefined rules covering general search strategies and rules for accessing specific sites.
• ‌Real-Time Progress Tracking‌: Supports CLI, TUI, and web-based workbenches; allows you to view completion progress, task flows, and frame-by-frame evidence; and supports resuming after an interruption.

SearchOS's Core Technology

• ‌Relational Search Abstraction‌: Define information retrieval tasks as the filling of structured relational models to unify the logical framework across various types of retrieval tasks.
• ‌SOCM Shared State Infrastructure‌: Externalize the search state into four types of shared data structures to prevent the loss of facts caused by dialogue context compression.
• ‌Parallel Scheduling on a Production Line‌: By adopting a micro-batch approach, multiple agents execute search chains in staggered intervals, and new tasks are immediately assigned to idle slots, thereby improving resource utilization.
• ‌Three-Tier Middleware Intervention Mechanism‌: Implements context control, loop recognition, and evidence anchoring and storage at the model-tool interaction layer, without relying on prompt constraints.
• ‌Evolution of Dynamic Modes‌: Dynamically adjust the table structure based on the entities discovered to accommodate the diversity of real-world search scenarios.

Use Cases for SearchOS

• Complex data retrieval tasks: such as bulk collection of entity attributes and comparative research.
• Long-term open-domain information gathering: such as comprehensive industry-wide research and multi-source fact-checking tasks.
• High-precision traceability-based searches: Such as verifying academic materials and checking compliance information, which require that each conclusion can be traced back to its original source.
• Multi-agent parallel search scenarios: Large-scale information gathering tasks that require collaboration among multiple agents and the avoidance of redundant work.

SearchOS'sProject Address

SearchOS is an open-source project that developers can access and use through the following methods:

• Project website: https://antins-labs.github.io/SearchOS/
• GitHub repository: https://github.com/antins-labs/SearchOS
• arXiv Technical Paper: https://arxiv.org/pdf/2607.15257

How to useSearchOS?

1. ‌environmental preparation‌: Visit the GitHub repository, clone the project, and complete the configuration of the model provider and API key.
2. ‌Start Interaction‌: Choose any of the following interfaces to start the system: CLI, full-screen TUI, or Web Research Workbench.
3. ‌Enter a query‌: Enter a natural-language question in the search box, and the system will automatically identify the task type and construct a relational model.
4. ‌View Progress‌: Real-time monitoring of model construction, coverage map completion status, and the task execution flow of each agent.
5. ‌Review Results‌: After completing the task, review the structured output; each data entry includes a citation that can be traced back to the original web page.
6. ‌Interruption Recovery‌: Long-range tasks support resuming after being interrupted, and you can review historical search trajectories at any time.

Comparison of similar products