A business may be one thing, but its work quickly becomes many projects. Research, product, engineering, launch, and operations each develop their own goals, decisions, and open questions. As Agents accelerate the execution, the person running the work becomes its memory, search engine, and handoff layer.
Why Superself was rebuilt.
The first Superself was a place where AI conversations turned into documents — knowledge you could keep. It worked, and it taught us the limit: stored knowledge does not compound by itself. Documents piled up, but nothing kept a project's scope separate, told a confirmed decision from a draft, or handed the current state to the next session. The archive grew while every new piece of work still started from zero.
Meanwhile execution moved to Agents. What work now needs is not another place to store documents but a layer that operates project Context — so people and Agents read the same truth, and what one session learns is where the next session starts. That layer is SharedContext, and Superself was rebuilt around it.
Execution got faster. Context became the bottleneck.
An Agent can research, draft, code, and analyze in minutes. But if every session must rediscover the goal, reconcile stale decisions, collect state from scattered documents, and ask a person to rebuild the briefing, the manual work has not disappeared — it now runs before every session.
Re-briefing every session is the immediate symptom. The wider problem is that several projects cannot keep advancing unless one person continuously routes their Context.
Isn't project Markdown already enough?
For some people, yes. A repository with a clear AGENTS.md, handoff notes, and Git history can be an excellent solution. Markdown is an open format, and SharedContext uses it too.
But generating a file is not the same as operating project Context. After the file exists, someone still has to keep each project's scope separate, tell a confirmed decision from an Agent proposal, point every session at the current version, and recover when a change is wrong. SharedContext supplies that operating layer without replacing Markdown.
What SharedContext does.
See the shared direction.
A Space carries Context that genuinely applies across its projects.
Keep execution scoped.
Each project has isolated Context, so work in one project does not silently become truth in another.
Read, work, and carry the result forward.
An Agent opens the selected project, follows its index, performs approved work, and preserves the changed state for the next reader.
People remain responsible for goals, trade-offs, scope, and completion. Agents can surface questions, propose next steps, execute approved work, and update Context. SharedContext is the first persistent layer that lets this loop survive beyond any single model, conversation, repository, or branch.
What shipped.
Drop 001 ships the Context machinery, not an autonomous project manager. The current product provides:
Remote, versioned Markdown
Durable documents with revision history and revertible changes.
Personal, Space, and Project scope
Context can follow one person, apply across a Space, or remain isolated to one project.
Indexes, list, search, and read
An Agent can start from a small index and load detail only when the work needs it.
Agent access through MCP
Claude Code has a dedicated plugin path. Other tools can connect when they support authenticated remote MCP; exact setup and validation vary by client.
What the first dogfood proved — and did not.
Superself first tested the layer on its own launch project. A new MCP session resolved the selected project, returned its scoped knowledge and working discipline, and kept subsequent session logs attached to that project. A project rename also remained resolvable on the next open. That is evidence that project identity and Context can survive beyond one chat.
It is not yet proof that every supported client completes the same handoff, that an Agent will never promote an unconfirmed proposal into a decision, or that the right project is selected automatically. Those claims require client-by-client and end-to-end validation, so they are not part of this release promise.
Where it fits — and where it does not.
If all of your work stays in one repository, you use a small fixed set of Agents, and maintaining project Markdown yourself is cheap, keep doing that. SharedContext is for the point where projects, sessions, and AI surfaces multiply enough that manual routing becomes part of the work.
SharedContext does not remember every conversation, make unconfirmed choices authoritative, coordinate project priorities on its own, or automatically choose the correct project after installation. Version history records how a document changed; it does not decide who had authority to confirm its content.
Coming from the previous Superself?
Your documents came along. Each space from the previous Superself was migrated into the new structure — related documents grouped into projects, each with a description and a narrative index an Agent can read. Sign in and your knowledge is already organized as SharedContext, ready to be the starting material for whatever you work on next. What changed feature by feature is in Updates.
Start with one real handoff.
Learn the Space → Project → Context model, then follow the first handoff guide. If the model fits your work, connect an Agent and verify the first session_open against a project you actually intend to continue.