antins-labs/SearchOS

Schedule search agents the way an OS schedules processes.

What it solves

SearchOS addresses common failures in long-horizon AI search tasks, such as opaque processes where facts are lost in conversation history, repetitive looping where agents re-search the same data, blurred roles where agents struggle to search and summarize simultaneously, and inability to access data behind anti-bot walls or login gates.

How it works

SearchOS operates as a multi-agent collaboration system that treats search state as a system asset rather than conversation history. It uses Search-Oriented Context Management (SOCM) to maintain a shared persistent state consisting of a task queue, an evidence graph, and a coverage map.

The process follows six steps:

  1. Explore: A scout agent identifies query types, hub pages, and candidate entities.
  2. Schema: The system builds a normalized coverage map (entity x attribute tables).
  3. Dispatch: Gaps in the map are split into subtasks and dispatched to parallel search agents.
  4. Extract: Extraction middleware automatically pulls structured data (entity, attribute, value, source) into the evidence graph.
  5. Assess: The system polls subtasks, handles conflicts, and targets remaining empty cells.
  6. Synthesize: A final answer is generated based on the completed coverage map with full citations.

Who it’s for

It is designed for users performing open-domain information seeking and deep research that requires structured, citation-grounded results across multiple entities and attributes.

Highlights

  • SOCM (Search-Oriented Context Management): Keeps task queues and evidence graphs persistent, allowing for session snapshots, restores, and replays.
  • Coverage-Map-Driven: Normalizes questions into tables to ensure every required data point is targeted until filled.
  • Pipelined-Parallel Agents: Overlaps search, open, and find stages to reduce total wall-clock time.
  • Citation-Grounded: Every extracted value is anchored to a source in the evidence graph.
  • Loop-Breaking Sensors: Employs five types of detection to prevent agents from repeating the same failed search queries.
  • Skill System: Includes specialized access skills to bypass anti-bot walls and strategy skills for multi-hop or ranking questions.

Related

  • Project
    MemTensor/MemOS

    MemOS 2.0 is a Memory Operating System for LLM‑based AI agents. It offers a unified API to store, retrieve, edit, and delete multi‑modal memories (text, images, tool traces) organized as a graph. The system can run as a hosted cloud service (API‑key access) or be self‑hosted via Docker (Neo4j + Qdrant) or a local SQLite plugin. Ready‑made plugins integrate MemOS with OpenClaw, Hermes, and DeepSeek Harness. Benchmarks show it outperforms existing commercial memory products. Licensed under Apache 2.0.

  • Project
    Narcooo/inkos

    An AI Agent system for long-form story creation and translation that manages world-building, character continuity, and narrative state to prevent plot holes and AI-style prose.

  • Project
    beilusaiying/always-accompany

    A multi-purpose AI agent framework focused on context and attention mechanisms, providing layered memory and dynamic retrieval for long-term companionship, coding, and work.

  • Project
    benchen4395/alpha_agentic_search

    A stable, low-latency multi-agent search framework featuring a 5-layer hierarchical memory RAG stack that allows the system to improve its performance as it is used.

  • Project
    kyegomez/swarms

    A production-ready multi-agent orchestration framework that provides pre-built architectures for coordinating multiple AI agents to solve complex tasks.