Enovari compared
Four systems, one table.
Three full comparisons behind it.
Each comparison is built around the one question where Enovari and the other system disagree. The table below lines those answers up. Where a page does not compare a row, the cell says so and sends you to the page instead of guessing.
| The question | Enovarithis product | Mem0vs Mem0 → | Zepvs Zep → | Lettavs Letta → |
|---|---|---|---|---|
| Where the memory lives | A SQLite file in a folder you own, plus a separate file for each persona. | The managed platform is its recommended path; a self-hosted Apache 2.0 build exists. | Not compared — see page | A git repository that belongs to the agent, hosted by Letta for cloud agents. |
| Who decides what is stored | Whoever makes the write call — a person, or the agent doing the work. | By default, an LLM extracts what to keep. infer=False stores raw content as given. |
Extracted facts are compared to existing edges; duplicates merge. | The agent, editing its own memory files; every edit is committed to git. |
| When two facts disagree | Both stay in force. A contradiction row stays unresolved until someone settles it with a written reason. | The older memory is marked superseded — not deleted, still returned, badged as history. | The older fact is invalidated, and kept with timestamps for the window it was true in. | Not compared — see page |
| Who runs the agent | The assistant you already use. Enovari runs no agent and ships no model. | Not compared — see page | Not compared — see page | Letta. You build a Letta agent and Letta runs it. |
| How your assistant reaches it | An MCP server on your own machine, over localhost. By default a browser-tab assistant cannot reach it. | A hosted MCP server over HTTPS, plus REST and SDKs. | Not compared — see page | The Agent SDK (TypeScript only) or ACP for editors. |
| Published benchmark numbers | None published. | Publishes its own, with its own caveats attached. | Not compared — see page | Not compared — see page |
Every cell summarises a claim its linked comparison page already makes, where the source for each is quoted and dated. The Mem0, Zep and Letta claims on those pages were re-checked 2026-09-14. No cell is a score.
- vs Mem0 Mem0 extracts what to remember. Enovari is told. Who decides what gets remembered, and whose computer the answer lands on. Read the full comparison →
- vs Zep Zep invalidates the older fact. Enovari leaves both open until someone writes why. What happens when two stored facts disagree, and the cost of our own choice. Read the full comparison →
- vs Letta Letta gives you an agent. Enovari gives yours a memory. Who runs the agent, and the Letta capabilities we could not honestly claim as missing. Read the full comparison →