Compare / Category
Copilot memories are written by the agent and cannot leave GitHub; Zephr memories are signed by a human and work everywhere. That is the structural moat — Zephr is the trusted continuity and evidence control plane across agent clients, and it does not compete as another memory API or another IDE. The useful comparison is rarely feature-for-feature; each page below opens with where the competitor is genuinely stronger, then explains the specific workflows where Zephr adds something.
These are the four capabilities that actually separate the categories, rather than the four that Zephr happens to win. Our own row reads partial on abstention, because retrieval abstains today but the provenance claim-support gate is not evaluated yet.
| Product | Cross-tool | Signed handoff | Abstention | Public MCP |
|---|---|---|---|---|
| Zephr | yes | yes | partial | yes |
| Mem0 | yes | no | no | yes |
| Generic memory API | partial | no | no | partial |
Which is better depends on what you measure. Switch the lens — the matrix re-reads the same products through continuity, evidence, or portability. Verdicts change because the question changed.
| Continuity | Cross-tool | Signed handoff | Abstention | Public MCP |
|---|---|---|---|---|
| Zephrthis page | ●supported | ●supported | ◐partial | ●supported |
| Mem0 | ●supported | ○absent | ○absent | ●supported |
| Generic memory API | ◐partial | ○absent | ○absent | ◐partial |
Verdicts sourced from ADR-Z-16 and public product documentation, read 2026-07-24. Cross-tool = the memory survives an editor switch; Signed handoff = the packet carries a verifiable signature.
verdicts are testable and dated — no scores, no winners
Every one of these products is good at something Zephr is not. Each page states that first, because a comparison that only lists your own wins tells the reader nothing they can act on.
Drop-in memory API — OSS + managed platform.
Where they win: Faster to adopt, simpler to call, and far more widely distributed.
Temporal graph (Graphiti OSS) + Zep Cloud.
Where they win: Temporal validity windows and a managed cloud path for production graphs.
Editor-native memory, deep IDE integration.
Where they win: Simpler and better integrated if you never leave one editor.
Stateful agent OS with memory inside the runtime.
Where they win: Better if you want an agent operating system, not a continuity layer.
Institutional knowledge graph over your corpus.
Where they win: Stronger graph-memory primitives for a single knowledge base.
Claims are sourced from ADR-Z-16 and a public-positioning re-check dated August 2026. Every product here is actively developed, so treat these as a dated snapshot rather than a forecast. Letta is a stateful agent runtime, not a memory API — a different category, so there is no vs-Letta page. If you need the agent to own its memory budget, use Letta; if you need a portable evidence layer under tools you already run, that is Zephr.
No product here wins every use case. Pick the one whose strengths match the workflow you actually have, not the one whose comparison page you happen to be reading.
The framework derives from the ADR-Z-16 competitive differentiation table. Last reviewed 2026-07-24.
you want fast onboarding, a global user memory, and broad MCP/plugin distribution — and provenance is not a hard requirement.
you need a temporal knowledge graph (Graphiti) or Zep Cloud governance — and coding-scoped handoff is secondary.
you live inside one editor, value deep IDE integration, and do not switch clients or need portable provenance.
you work across multiple agents, need per-worktree and per-session scope, and want authorization plus receipts before memory is ever retrieved.
zephr recallThe one command that makes the category claim checkable — recall runs against your own admitted worktrees.
Read how Zephr installs, connects, and enforces trust — or go straight to the trust and security model that the comparisons keep pointing at.