Cursor2d ago
Last month’s session
Tool-local history.
One memory behind every agent and every tool.
Your team does not use one assistant. What an agent learns in Cursor is invisible to Copilot, and starts from nothing in Claude. Virgo is a single shared memory reachable over MCP, so context follows the work, not the vendor.
sourcesPR #4720#securityCursor session60s TTL
›
⏎ send · esc to interrupt
Tool-local memory fragments
The decision exists. It lives in one IDE’s history on one machine, invisible to every other tool your team opens today.
Not searchable
Out of reach
ClaudeTue
A teammate’s session
Lives on their machine.
Starts fresh
Copilotnow
Inline suggestions
No session memory.
Where did we settle the caching rule?
“It was in some session…”
Never posted
SlackMay
#eng
The decision came after the thread.
Stale
Google DocsApr
Your own notes
Three repos ago.
Unlinked
GitHub2d ago
PR #4720
The comment thread knows.
One memory, reachable from every tool
Commit from Cursor, read from Claude an hour later. The memory lives under the tools, over MCP and a plain REST API, so adding a tool never resets the team.
One memory pays off in every tool
Committed once, a decision follows the work into whichever tool opens next.
For engineers
Switch tools without starting over.
The caching rule settled in a Cursor session is waiting in Claude an hour later, with the incident behind it attached. Your context follows the work, not the vendor.
- The decision: per-user keys, 60-second TTL.
- The reason: the cross-user leak that settled it.
- The reach: any tool that speaks MCP or REST.
For AI agents
A standard protocol, not a walled garden.
Virgo connects through MCP and a plain REST API, so every agent reads and writes the same memory. Add a new coding tool next year and it inherits everything the team taught the old ones.
3xfewer iterations
Roughly 3x fewer iterations in code generation when the relevant memory is on hand instead of rediscovered.
Built on Agent memory → and Save what your agents learn →