Search2o documentation

Search2o is a platform for creating and running AI agents behind a search interface. Developers describe agents in JSON with Python expressions. Users type what they need, the matching agent runs, and follow-ups continue the same conversation.

Start here

• IntroductionWhat Search2o is, the three parts of the system, and how a request becomes a result.
• Getting startedCreate your account, install the agent server with pip, run it with your license, and build a first agent.
• Agent definitionThe JSON structure, dynamic strings, functions, commands and variables.
• Command referenceAll 23 commands, every field, with working examples.

Building and running agents

• Agent executionOutput streaming, order of execution and conversation state.
• Development processDrafts, the editor, Draft with AI, trace and validation, and publishing.
• SearchHow to describe an agent so users can find it, how matching behaves, tags, settings, how good search is, and search from an interface of your own.

Operating the system

• Agent runtimeWhat an agent server builds from the configuration and runs agents against: the allowlist, system variables, compile rules and runtime limits, connection pools, and how a change reaches every server.
• ProfilesThe LLM, prompt, API, database and MCP profiles that agents refer to by name.
• ReportsAgent performance, errors, LLM cost and user usage.
• System managementUsers and roles, service accounts, agent servers and how to size them, authentication, single sign-on and notifications.
• Security and privacyLicense key rotation, the secret vault, encryption and key rotation, where every kind of data ends up, and running several environments.
• GUIHow the GUI runs, and a tour of every page in the bundled interface, by role.

Reference and more

• REST APISign in, run agents, stream, search, manage conversations, automate tasks with an LLM, and build your own client on the same API the GUI uses.
• Chat integrationsLet people use agents from Slack, Microsoft Teams or Google Chat: connecting a person, finding and running an agent, and prompts that write most of the bot.
• Skill integrationLet an Agent Skills client such as Claude Code run your published agents: how the skill works, installing it, and the notes for the team that rolls it out.
• LLMThe three bundled vendors, connecting to any other LLM, and running code that an LLM writes.
• Support and licensingAsking a question of the documentation, how to reach Search2o support, the license of the agent server, and billing.