The AI agent space just got a new player with the release of Search2o, a stateless, asynchronous Python server now available in open beta. The project offers a REST API and a bundled GUI for running agents directly within your organization's infrastructure. Unlike many cloud-locked competitors, Search2o emphasizes local control and searchability, allowing developers to build, validate, and publish task-specific agents that can be queried via plain English.

Architecture and Setup

Installation is straightforward for Python developers: pip install search2o gets you the core server. The architecture is designed to be flexible regarding LLM backends. It supports self-hosted models via Ollama, vLLM, or SGLang without requiring vendor API keys, while also accommodating OpenAI, Anthropic, and Gemini through standard environment variables. The server runs locally at http://127.0.0.1:9020, providing a dedicated UI for user management and agent creation. Notably, the GUI is currently bundled with the server binary, though the GitHub repository indicates it will eventually be split into a separate project.

The JSONC DSL and Workflow

The core innovation here is the use of a JSONC (JSON with Comments) DSL for defining agent behavior. The workflow involves creating an LLM profile, drafting agents through an AI-assisted interface, and then validating and publishing them. Once published, these agents are indexed for search. Users can type a question into a search box, and the system matches it to the appropriate agent, which then executes and returns an answer. This search-first approach to agent invocation aims to reduce the friction of managing multiple specialized agents by treating them as searchable resources rather than static endpoints.

Licensing and Integration

Search2o is not open source in the traditional sense; it is source-available and proprietary, governed by the Search2o Software License Agreement. External code contributions and pull requests are currently not accepted, though bug reports and feature requests are welcomed. For developers already entrenched in the agentic workflow, the system integrates with Claude Code via the search2o-skill, allowing the same agents to be utilized outside the web GUI. The current release, version 0.11.0, was uploaded to PyPI on October 7, 2026, with both source and wheel distributions totaling 8.0 MB.

Key Takeaways

  • Search2o offers a local-first, stateless Python server for AI agents, supporting both self-hosted and API-based LLMs.
  • Agents are defined using a JSONC DSL and are made discoverable through a natural language search interface.
  • The project is in open beta and is source-available but proprietary, with no external contributions currently accepted.
  • Integration with Claude Code via search2o-skill extends its utility beyond the standalone web GUI.

The Bottom Line

Search2oโ€™s local-first approach is a breath of fresh air, but the proprietary, closed-contribution model risks limiting community-driven innovation in the fast-moving agent ecosystem.