SharedContext makes each project's durable knowledge available across Agents, models, repositories, branches, and sessions without merging every project into one memory.
The problem
As projects multiply, important work is often trapped in old conversations and scattered files. A decision may have been made, a constraint discovered, or a release state clarified, yet the next Agent begins without it and asks the user to reconstruct the briefing. The person running the work becomes the memory and router for every project.
How it works
SharedContext separates working memory from durable knowledge.
- A session map records completed steps, decisions, assumptions, and open questions while work is active.
- Versioned Markdown documents preserve knowledge that should outlive the session.
- Scope separates project-specific, Space-shared, and personal Context. Project Context is isolated between sibling projects.
- Index files and search give an Agent a path to the right document without loading the whole corpus into one prompt.
The value
The next session starts further ahead. It can recover after compaction, find an earlier decision before recreating it, and update the selected project's shared understanding when circumstances change.
SharedContext does not try to remember every token, automatically choose the correct project, or decide whether an Agent proposal is authoritative. It preserves what future work needs and keeps revision history; people still confirm goals, scope, priorities, trade-offs, and completion.