Each editor remembers only itself
- Claude Codelocked
- Cursorlocked
- Copilotlocked
Switching tools means starting from a different — and often stale — memory, with no way to tell which of the three is right.
3 stores · 0 shared
Compare / vs Copilot Memory
Neutral continuityvs editor-native memory.
Copilot Memory is deeply integrated with the IDE, and inside that IDE it is the simpler choice. Zephr sits above the editor: one continuity plane every agent can read and write, with provenance attached and policy enforced. The comparison is architectural, so the right answer depends mostly on how many tools you actually use.
Snapshot · July 2026ADR-Z-16sourced 2026-07-24
Silo vs plane
One editor, one memory. Many editors, still one memory.
Copilot Memory keeps context inside the IDE. Zephr keeps it in a neutral layer that every IDE can read and write. If you only ever open one editor, the first row is not a limitation.
The same three editors, two architectures
On the left each editor holds its own memory and cannot see the others. On the right the same editors read one provenance-backed ledger, with authorization and receipts at every retrieval.
One ledger, every agent
- Claude Codeconnected
- Cursorconnected
- Copilotconnected
One provenance-backed memory shared across connected agents, with authorization and a receipt on every retrieval.
1 store · 3 connected
Illustrative example — not live data. Real records live in the local evidence graph, not on this page.
Where Copilot Memory wins
Copilot Memory is simpler inside one editor.
If you use Copilot and never switch editors, Copilot Memory is the straightforward choice and Zephr is extra machinery for a problem you do not have. Zephr earns its place when the workflow spans tools.
Deep integration
Native access to editor and GitHub ecosystem context that an external layer cannot see, and does not try to.
Frictionless capture
Nothing to install or admit. Inside a single IDE workflow the memory is simply there, which is a real advantage.
Repo-state revalidation
Validated facts are re-checked against the repository, so stale context is caught rather than served indefinitely.
Simplicity
If you never switch tools, a neutral portability layer is solving a problem you do not have.
Zephr's edge
Where Zephr adds value.
Zephr is not a better Copilot Memory. It is a neutral portability and trust layer that outlives any single editor — which only matters once you have more than one.
Survives IDE switches
A neutral layer above the editor, so moving from one agent to another mid-task carries the decision instead of resetting it.
MCP-native
Works across Claude Code, Cursor, Copilot, Windsurf, and custom agents through one contract rather than per-editor integrations.
Provenance and authorization
A full provenance chain per claim, and authorization evaluated before retrieval rather than trusted by proximity.
Rules enforced, not advised
Project policy is evaluated by the Trust Firewall on every request, instead of living in a file each tool may or may not read.
Decision framework
Choose Copilot Memory or Zephr?
The decision is mostly about how many tools you use, and how much you need to prove where an answer came from.
Choose Copilot Memory if…
You stay inside one editor, value deep IDE integration, and do not need cross-client portability or a provenance chain behind every retrieval.
Choose Zephr if…
You move between Claude Code, Cursor, Copilot, or custom agents, and you want project-scoped memory with enforced policy and evidence receipts that follow you across them.
See the trust model.
The trust and security page explains authorization before retrieval, the provenance chain, and honest abstention — the mechanics behind the right-hand column above.