SF RecallJoin the waitlist
MODEL-AGNOSTIC MEMORY FOR AI AGENTS
Claude has a memory. It is Claude's. SF Recall mounts over MCP — so any agent that speaks the protocol reads and writes the same standups, decisions and unfinished work. Change model, change tool, wipe the laptop; the memory is still yours. A daily cycle that verifies work before anyone builds it, and a review rubric that compounds, are built on top.
PROVEN BY WHAT IT BUILT
SaaS Factory is a platform where AI agents build and operate real software products — planning, coding, releasing and running them for months on end. These are the tools that ran it: the cycle that stopped it re-building what already existed, the sweeps that kept what it shipped honest, and the memory both stand on. SF Recall is that working set, brought to the world.
See what it built — saas-factory.aiTHE TOOLKIT
Work gathered from wherever it arrives, checked against the real code before anyone builds it, and a board a human decides from — once a day.
A sweep of what you already built, judged by a rubric that keeps every finding type it has ever learned about your product — and grows on every pass.
BUILT FOR DROP-IN
The memory is reached over an open protocol with standard OAuth, so there is no per-vendor integration to wait for and nothing in it is shaped around one model.
Any MCP-capable agent Dynamic client registration and OAuth 2.0 with PKCE. An agent registers itself and gets the whole surface — standup, recall, deposit, the board.
Or no agent at all There is a REST surface too. If you are building something agentic of your own, you can give it a memory instead of building one.
Generic by design Deposit, recall, supersede, standup, attention items. None of it assumes you are writing code — the development tools are built ON the memory, not baked into it.
Standup, decisions, handover — shared across the team and surviving a wiped laptop. The substrate the other two tools run on.