AI builders' handbook
This handbook is for the people who build with Labtagon AI OS: they connect Claude Code, Codex, Cursor or claude.ai, decide what each tool may use, build tool servers and skills with their assistant, set up agents that work on their own, and call AI OS from their own applications. Most of it happens in Studio.
In Studio’s sidebar you find your agents, the button New agent with More to make beside it (New agent with AI, New server, New skill), and Tool servers, Skills and Knowledge. Pages such as Profiles, Connections, API keys, Jobs and Your sign-ins you find with the sidebar’s search. Whether you may build is your organisation’s decision (the permission Build).
Connect
Section titled “Connect”Clients and the gatewayOne address for every tool, how clients sign in, and what the gateway does with each call.
Connecting clientsThe exact steps for Claude Code, Codex, Cursor, claude.ai and AI OS Workbench.
ProfilesWhat a client may use: tool servers, tools, sign-ins, skills, capabilities.
Building a tool serverDescribe it, your assistant writes and builds it, checks run, you approve, versions.
SkillsWritten procedures your clients and agents read and follow.
Agents and knowledge
Section titled “Agents and knowledge”Agents and jobsAgents that work by hand, on a schedule or when a webhook calls them.
Memory spacesMemories and wikis, shared with people and groups.
Your own application
Section titled “Your own application”AI OS from your own applicationThe OpenAI-compatible API with keys and budgets, and pipelines that bring your tools along.