Sno Station CLI
One local memory for Codex, Claude Code, Cursor, Hermes Agent and OpenClaw: agents hand off tasks, review each other and refine their skills overnight
- 765 ★ on the repository
- Last commit
- Apache-2.0
Broad access
Why this level
- Hooks go into every agent found on the computer unless you name one with the `--harness` flag
- Memory stores the content of your agent sessions, and in REM Enhanced mode the memory text goes to Sno's cloud
- The nightly loop proposes edits to agent skills, and they take effect once you accept them
Install
- Works in
Manual install
sh -c 'sno_installer_body=$(curl -fsSL https://sno.ai/install) && printf "%s\n" "$sno_installer_body" | sh' && ~/.local/bin/sno setupInstalls `sno`, then Sno Station, and connects every agent found. No account or API key needed. Then say «Sno onboard» in an agent chat.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- Hooks go into every agent found on the computer unless you name one with the `--harness` flag
- Memory stores the content of your agent sessions, and in REM Enhanced mode the memory text goes to Sno's cloud
- The nightly loop proposes edits to agent skills, and they take effect once you accept them
This is third-party code. Review the repository files before installing.
What it does
Sno Station turns the agents you already use into a squad. Codex, Claude Code, Cursor, Hermes Agent and OpenClaw read and write one encrypted memory on your computer. The encryption key never leaves the machine, and no account, server or background daemon is required. When one agent hits its limit, you say `sno reach call` from another one. It picks the task up from the shared memory and the mailbox with the context intact. For a second opinion, ask any agent to run `peer-review` on another agent's work: the reviewer is always a different agent, so it has no habit of agreeing with the first. Every night a REM cycle runs. It strengthens the memory across all sessions at once, retires facts that are no longer true, reads the day's sessions, finds mistakes that repeat and proposes changes to the agents' own skills. In the morning you accept the ones you like with `sno rem-reflect accept <id>`. Nothing changes without that.
Who it is for. For developers who run two or more agents side by side and move results between them by hand.
Good fit when
- Claude Code hit its limit mid-task and Codex can carry on from the same place without copying context
- You need an independent review: Codex checks Claude Code's work or the other way round
- The same agent mistakes keep returning session after session and you want them turned into skill rules
- Memory should sit on your disk encrypted, with no cloud and no sign-up
- Cursor in the app chat or in cursor-agent should remember decisions made in other agents
Not a fit when
- You work with a single agent: task hand-over and cross-review will not help
- You need Windows: Cursor and Sno Reach support is described for Linux and macOS
- You need Cursor's cloud agents: they are not supported
Example request
Sno onboardLimitations
Cursor joins through the app chat on macOS and the cursor-agent command line on Linux and macOS. In the app, the first chat of a new window gets its repository brief with the first prompt, not before it. cursor-agent -p gets no per-prompt memory. Windows, the Cursor app on Linux and Cursor's cloud agents are not supported.
The project's status table marks a row proven only after a run on a clean machine. The Cursor app on macOS was still being proven when this entry was written.
In REM Enhanced mode, your memory text and the lessons your agents learn go to your own space in Sno's cloud and are processed there for you alone. The cloud side is optional.
How to disable. All `sno` commands are listed in docs/sno-commands.md in the repository. Nightly proposals apply only after `sno rem-reflect accept`, so declining is just not accepting.
FAQ
How do I install Sno Station for Claude Code and Codex?
Run the install command from the README: it fetches the installer from sno.ai, installs `sno`, then Sno Station, and connects the agents it finds on the computer. To connect just one, add `--harness claude` or `--harness codex` to `sno setup`. Then say «Sno onboard» in any agent chat.
What do I do when an agent runs out of quota?
Say `sno reach call` from another agent. It takes the task from the shared memory and the mailbox with the same context. In the repository's example, the first agent wrote a note on what was done and half-done plus the exact commit to continue from, and the second picked up at the first unfinished task.
Do I need a sno.ai account?
No, everything works without one. An account is only for the sno.ai dashboard: what the reviews caught, what the agents learned overnight, which skills improved. `sno account claim` links the computer to an existing account.
What goes to the cloud in REM Enhanced mode?
Your memory text and the lessons your agents learn. They land in your own space in Sno's cloud and a Sno program processes them for you alone. REM runs at most once a day for each part of the memory: usually overnight, or earlier once about 100 new memories pile up.
How is Sno Station different from claude-mem?
claude-mem keeps memory for Claude Code. Sno Station holds one memory across five agents and adds task hand-over, cross-review and a nightly loop that proposes skill edits.
Discussion
No questions or reviews yet.
Related
CLI toolsKnowledge-graph memory for agents: Claude Code and Codex plugins, an MCP server and a Python library
Shared local memory for AI agents: Claude Code, Codex, Cursor and others find the agreements of past sessions in it and record new ones
Local memory for agents: stores conversations and project files verbatim and retrieves them with semantic search over MCP
Local persistent memory for coding agents: hooks record every session, and Claude Code, Codex or Cursor recall past decisions through 54 MCP tools
Long-term memory for agents: a remote MCP server, plugins for Claude Code, Cursor and Codex, and an API for your own apps
A Claude Code skill that pulls memory from Memori Cloud before every non-trivial answer and writes new facts back to it
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| Sno Station (this entry) | no | 765 | |
| Claude-Mem | no | 102,227 | memory for Claude Code only, with no task hand-over or nightly loop |
| Claude-Codex Review Skill | no | 6 | a single skill for plan or code review by a second agent through files |
| Claude Squad | no | 8,645 | runs parallel agent sessions in tmux and git worktrees |
README in short
The project has been public since 18 September 2026 under Apache-2.0. The repository holds a memory package per agent (mem-claude, mem-codex, mem-cursor, mem-hermes, mem-claw), the Sno Reach utility that lets agents message each other without a daemon. The status table marks a row proven only after a clean-machine run, and the real-world sections come with screenshots and logs. A separate section explains why the memory forgets on purpose: a stale fact such as a cancelled preference does more harm than a lost one.