1In short
Morion is a desktop app with notes and a task board that Claude Code, Codex, Cursor and any MCP client share. You and your agents write notes and tickets; Mo indexes them into projects and topics, answers with the source and a status on every claim, and hands the agent that picks up a ticket everything related to it.
claude-mem is an open-source plugin with about 95,000 GitHub stars: a second model watches your agent’s tool calls, compresses them into observations and injects the relevant ones into later sessions, with no writing on your side, in Claude Code, Codex and Cursor.
Pick Morion if you want a project memory you can read and correct, and a board your agents work from. Keep claude-mem for session recall with zero effort; the two work side by side.
2Where claude-mem costs you
The same design that makes capture effortless has a cost, and the project’s own GitHub issues describe it.
- A second model on every tool call. The observer runs alongside your agent, so memory costs tokens or money all day. One issue reports $17 in three hours of observation, another 40% of a session’s tokens gone at start.
- A stack to maintain. Node, Bun, uv, a background worker and a vector store. Issues report runaway processes, a database that grew to 136 GB, and updates that broke installs.
- What the agent did, not what you decided. Observations are compressed summaries of actions. One review notes that nuance gets lost in compression, and nothing marks a claim as decided, provisional or overridden.
- Forward only. It records from the day you install it; the notes and specs you already have are not part of its memory.
- Data leaving the machine. Telemetry is on by default since June, and cloud sync, when you turn it on, uploads observations and prompt text.
3Side by side
| claude-mem | Morion | |
|---|---|---|
| Memory captured from every session automatically | ||
| Notes you write and edit | — | |
| Full-text and semantic search | ||
| Answers that cite their source | ||
| Status per claim, conflicts and gaps | — | |
| Works with Claude Code, Codex and Cursor | ||
| Task board and work packets | — | |
| Workflow runner with a human gate | — | |
| Folder and per-note permissions for agents | ||
| Desktop app, nothing else to install | — | |
| Open source | ||
| Sync across machines and team memory | — |
has it, has part of it, — does not. As of September 2026. Morion does not record sessions; agents write notes and ticket comments over MCP when their instructions say so. claude-mem cites observation IDs; Morion cites the note a claim came from. claude-mem’s private tags keep parts of a session out of memory, not whole folders. claude-mem’s sync and team memory are paid; Morion’s sync is announced for Max and not shipped. For Morion, 2.0 has a Free plan and is not open source; the earlier 1.5 line, Morion Community, is open source and free but frozen.
4A log or a record of decisions
Most agent memory tools solve storage: keep everything, find the relevant piece later. The harder part is curation: which of the stored things is still true.
claude-mem stores what happened. If you chose one approach on Monday and replaced it on Thursday, both are in the log, and the agent has to guess which one holds.
Morion stores what you and your agents wrote down as notes: the spec, the decision, the ticket and its comments. Mo reads them into projects and topics and answers with the note each claim came from and a status: decided, provisional, superseded. When two notes disagree, the answer says so; when nothing covers the question, it says that too, instead of filling the gap.
Yes, mostly. Reminders are scheduled locally and mirrored to the server queue on sync Reminder architecture decided. Push delivery is out of scope until v2 Reminder architecture decided.
Whether queued deletes replay in order is still open Queue writes while offline · comments provisional.
Conflict: the March spec says reminders are server-first Sync protocol v1 superseded; the decision note overrides it.
Gap: nothing in the notes covers reminder time zones. Want me to open a ticket for it?
It also reaches back. Morion indexes the notes you already have, imported as Markdown, so the first answer can draw on a year of decisions instead of starting on the day you install it.
5Which one fits you
if you work alone in one repository, want the agent to pick up where the last session stopped, and do not want to write anything down.
if the project has decisions, specs and a plan that outlive any session, if two agents from different vendors work on it, or if you want tickets your agents claim and a loop where one writes and another reviews.
if you want session recall and a project record. Both connect to Claude Code at the same time: claude-mem answers “what did we do last time”, Morion answers “what did we decide, and what is next”.
6Moving over
Install Morion and connect Claude Code (and Codex, if you use it) in Settings → Integrations → Apps and Agents; one click writes the MCP entry, and the agent skill installs from Settings → Skills. Import the Markdown you already keep: CLAUDE.md files, the ones claude-mem generated, specs and decision notes. Turn on AI access for the folder and Mo indexes them into topics.
From then on, ask the agent to write decisions into Morion instead of relying on the log. The walk-through: From CLAUDE.md to a memory that cites.
7Questions
The plugin is open source under Apache 2.0 and free to run on your own model access. CMEM Pro is a paid tier for hosted processing and sync, and Team Cloud is priced per seat.
Yes. Since May 2026 claude-mem has Codex hooks, and since September plugins for Cursor, alongside OpenCode and Antigravity.
No. Morion does not watch every tool call. Your agents write what matters into notes and ticket comments over MCP, you write your own notes, and Mo indexes all of it into projects and topics.
Yes. They do different jobs: claude-mem recalls what happened in past sessions, Morion keeps decisions, specs and tickets that every agent reads. Both connect to Claude Code at the same time, and Morion has a Free plan.
Its telemetry is on by default since June 2026, with an opt-out. Cloud sync is off unless you turn it on; then it uploads observation narratives and prompt text to cmem.ai.
Sources: the claude-mem repository, README, releases and issues, docs.claude-mem.ai and cmem.ai, checked on 25 September 2026. See also Mem0 alternative and Supermemory alternative.