Open-source developer preview
Work with AI.
Not just through it.
GoTo Agents is a collaboration workbench for humans and AI agents. AI brings expert breadth on every subject. You bring the context, the taste, and the ideas that arrive overnight. This is a workspace built for that loop.
Projects, agents, skills, reference docs, and shared memory — all plain markdown on your machine, with any model provider you choose. Pick up tomorrow exactly where you left off today.
THE PREMISE
The industry is racing to remove the human. We think that's the wrong race.
The current trend is autonomy: swarms of agents in loops, doing everything themselves, judged by how little you touch them. That works for some tasks. But the work that matters most — creative work, thinking work, work where the outcome should be yours — doesn't improve by removing you from it.
Your best ideas don't arrive in a sixty-second loop. They arrive in the shower, on a walk, the next morning. When everyone has access to the same powerful models, the differentiation isn't the prompt — it's the judgment, taste, and lived context only a human collaborator brings. GoTo Agents is built to make that collaboration fast, low-friction, and durable across days.
THE LOOP
Iterate over days, not minutes — without losing the thread.
Work a session. Chat with an agent that knows your project, has the right skills loaded, and can read and write your actual files — lyrics, chapters, plans, designs, whatever the work is.
Everything worth keeping is kept.
Decisions land in shared project memory. Where things stand lives in the
project files themselves. The
reference library holds the handbook, the style guide, the source material.
Nothing lives only in a chat transcript.
Come back tomorrow with the better idea.
Every agent in the workspace picks up the same shared context — the memory,
the status, the library. No re-explaining. The overnight insight goes
straight into the work.
WHAT IT IS
GoTo Agents
A self-hosted workspace where agents are collaborators, not couriers.
Workspaces hold agents, skills, tools, and a shared reference library. Agents work with you on projects — real folders of real files — and share one memory of what you've decided together. Everything is inspectable, editable markdown. Nothing is hidden.
Workspaces & agents
Build focused agents for the work you actually do — a co-writer, an editor, a producer, a researcher — each with its own instructions, skills, and reference docs, all defined in markdown you can open and edit.
Shared memory, visible files
Agents share one project memory (MEMORY.md) and learn your
preferences in your profile — following the same conventions Claude and
other modern agents are trained on. What an agent believes is always one
click from view, and one edit from corrected.
Export & share workspaces
Package a workspace — its agents, skills, and reference library — as a portable zip, and import it somewhere else with one click. That's exactly how the Pulse Studio demo ships.
Agent Skills, the open standard
Skills follow the cross-vendor SKILL.md standard. Drop in skills from the ecosystem and they work; author skills here and take them anywhere.
Reference docs
Give a workspace its handbook, an agent its core references, and yourself a personal library — indexed for agents automatically, pulled in on demand or with an @-mention.
Local-first, provider-agnostic
Runs on your machine, works on your files, keeps your data. Run agents on a subscription you already pay for — Claude, ChatGPT — or point them at an API key instead, chosen per agent, with no separate metered bill required to get started.
MCP, both directions
Attach MCP servers to give agents new capabilities — or use GoTo Agents itself as an MCP server and drive it from Claude Code, Claude Desktop, or any MCP client you already use.
Reachable from anywhere
A local service you can talk to from its own web UI, from a Python client for scripts and notebooks, or from your own application over MCP — without stuffing agent config into every project or home folder.
SEE IT
From import to first draft in ten minutes.
WHO IT'S FOR
Built for work where the human is the point.
Creative projects. Songs, stories, scripts, worlds. An agent that knows your project's history, your influences, and yesterday's decisions — and never loses the plot between sessions.
Thinking and writing work. Research, plans, essays, product ideas. Iterate with an expert-on-everything that keeps the source material, the drafts, and the decisions organized in files you own.
Developers who want a workbench, not a framework. No DSL, no orchestration graphs. Markdown definitions, a local web UI, open standards, and a codebase you can read.
HOW IT'S DIFFERENT
Where the autonomy tools still fit.
GoTo Agents doesn't compete with coding-agent swarms.
If a task is fully specifiable — run the tests, fix the lint, scaffold the
app — automation loops do it well, and you should use them. GoTo Agents is
for the work you can't fully specify up front, because discovering
what you want is part of the work.
It's not another chat app either.
Chat apps forget. Here, conversation is just the surface; the substance
accumulates in project files, shared memory, and a reference library that
every agent in the workspace draws from — session after session, day after day.
Open standards over lock-in.
Agent Skills, MCP, markdown-with-frontmatter definitions, and memory
conventions shared with the wider ecosystem. Your content stays portable —
in both directions.
BUILD WITH IT
Agent-enable your own app — without building an agent host.
Build the app, then give it experts. A small MCP server in your application is often all it takes to let a GoTo Agents team read its data, propose changes, and ask before it guesses — instead of operating your app through a generic screen. Pulse Studio, GoTo Agents' own bundled demo, is a self-contained example: a three-agent research-and-drafting team you can import and run in minutes.
Real-world proof: SoundBench. SoundBench, an AI music production tool, ships its own built-in co-pilot and exposes an MCP server over its readings and files. Layer the O.H.R. Workspace on top — a GoTo Agents team of Bela, Styles, and Lena — for end-to-end expert collaboration, with every agent working through SoundBench's own tooling and projects.
Yours to build. The pattern is documented and a reference implementation is open to read. Deeper GoTo Agents-side tooling for app discovery and one-command workspace install is in active development.
Status
Open-source developer preview.
GoTo Agents is early software for builders and creators who want to shape what human–AI collaboration looks like. Self-hosted, Python + FastAPI, MIT-spirited, and readable in an afternoon.