Spindle
A local memory layer that helps AI tools carry useful context between sessions without mixing every project and preference together.
- Role
- Product design and engineering
- Disciplines
- Product design, Systems engineering, Browser and CLI interfaces
Context
Using several AI tools meant repeating the same setup over and over. I would explain the project again, attach the same files, and restate choices that had already taken hours to reach. The tools could help with the next step, but they did not reliably carry the work forward.
Insight
My first answer was to save more. That created a different problem. A useful project choice, a temporary instruction, and a personal preference should not all be remembered in the same way. Spindle became a system for deciding what gets saved, where it belongs, and when another tool should see it.
System
Spindle receives context from a browser or terminal, records where it came from, gives it a scope, and retrieves it when the current task calls for it. A memory can belong to one session, one project, one workspace, or an approved global layer.
How the connected parts share context
Browser, extension, and command-line clients use the same capture and retrieval contracts. Importers for past assistant histories and skill files enter through the same path, which keeps imported context from becoming a separate hidden system.
Decisions
- Make scope visible
A memory says if it belongs to the session, project, workspace, or global layer. The user should not have to guess why it appeared.
- Keep the source
Imported and automatically saved records keep where they came from, so they can be checked, corrected, or removed.
- Ask before saving globally
One tool seeing something once is not enough reason to make it available everywhere.
- Use one connection model
The browser, extension, terminal, and future clients connect through the same control API instead of growing separate shortcuts.
Experience
Reflection
The core memory, import, sync, extension, and command-line contracts exist. The remaining work is less interesting but more important: testing real providers, using Spindle across longer projects, and seeing where retrieval is wrong. This page describes the working system without treating unfinished testing as adoption.