GoTo Agents is a workspace for working with AI agents.
A self-hosted web app where you build agents that collaborate with you on real projects — creative work, writing, research, anything that unfolds over days. Agents share project memory, draw on a common reference library, and everything they know lives in plain markdown files you can open and edit.
Core concepts
-
Workspace — an isolated container for agents, skills,
tools, MCP servers, and a shared reference library. Defined by
WORKSPACE.md. -
Agent — an AI collaborator with its own instructions,
skills, tools, and reference docs. Defined by
AGENT.md(YAML frontmatter + markdown instructions) with an editable system prompt template. -
Skill — a capability loaded on demand, following the
cross-vendor Agent Skills standard: a directory with
SKILL.md, optionalscripts/andreferences/. Ecosystem skills drop in unchanged. -
Project — a real folder of real files that you and the
agents work on together. Each project carries its own instructions
(
GOTOAGENTS.md) and shared memory (.gotoagents/MEMORY.md). - Reference docs — the library: workspace-level handbooks, agent-level core references, and your personal documents. Indexed for agents automatically, read in full on demand.
-
Profile — who is at the keyboard. Your declared
preferences (
PROFILE.md) and what agents learn about you (MEMORY.md) follow you across projects. - Tools & MCP — file, shell, and task tools built in; MCP servers attach external capabilities. GoTo Agents is also an MCP server itself, so other MCP clients can drive it.
Philosophy
GoTo Agents deliberately isn't an autonomy framework. It's built on the view that for creative and thinking work, the human is the point: AI contributes expert breadth, you contribute context, taste, and the ideas that arrive overnight. The design follows from that:
- Continuity over throughput — memory, status, and the library make day-scale iteration seamless.
- Visible over hidden — every definition, memory, and prompt template is editable markdown, never buried state.
- Standards over lock-in — Agent Skills, MCP, and shared memory conventions keep your content portable both ways.
- Local-first — your machine, your files, your choice of model provider.
Ready to try it? See it work in the Quickstart, then build your own workspace in Getting Started.