Skip to content
Desktop only

AI agents and the silonotes command

Let an AI agent such as Claude Code read and add to the SILOs you choose, without ever deleting or overwriting your writing.

AI agents (Claude Code, other coding agents, or your own scripts) can work with your notes through a small command called silonotes. An agent can read the SILOs you allow, write new notes, add to existing ones, and tick to-dos. It cannot delete anything, it cannot change what you wrote, and it never sees your settings, keys, SILO Sam or the Library. It is off until you turn it on.

Turning it on

  1. 1Open Settings, Agent access and switch on Let AI agents use my notes. Read the warning and confirm with Turn on at my own risk.
  2. 2Under Which SILOs, choose Read or Read & write for each SILO an agent may use. A choice covers that SILO and every SILO inside it; everything else stays No access.
  3. 3Tell your agent to run silonotes help, and give it the AGENTS.md file that comes with the command. It explains the rules in the agent's own terms.

Read lets an agent list, read and search. Read & write also lets it create notes and sub-SILOs, add to notes, and tick to-dos. Switching the main toggle off makes the command refuse everything at once, whatever you granted.

Getting the command

The command comes with the desktop download as a small folder, silonotes-cli, holding the command and AGENTS.md. It runs with Node.js 22.5 or later (silonotes help), or with the SILONotes app's own runtime if you do not have Node (SILONOTES_APP=/path/to/SILONotes.AppImage silonotes help). It reads the same notebook as the app on that computer; nothing goes through a server.

What an agent can and cannot change

  • It never deletes. There is no delete command. An agent can take back only a block it wrote itself.
  • Your writing stays yours. An agent may edit or remove only blocks it wrote that nobody has changed since. Once you edit one of its blocks, that block is yours and the agent cannot touch it.
  • It signs its work. A note an agent creates carries the tag agent-<its name>. When it adds to a note it did not create, each visit starts with one line, "Added by <name>, <date and time>", followed by its blocks.
  • To-dos: it may tick any to-do, but untick only the ones it ticked.
  • Locked notes are never changed. Library books, SILO Sam, settings and API keys are out of its reach.
  • Limits: at most 30 changes a minute per agent, and 256 KB of text per command. Sub-SILOs it makes follow the same four-level limit as yours.

Seeing and undoing what it did

  • Before every change, the note's previous version is saved in its History as "Before an agent's change". Restore it from there like any checkpoint.
  • Settings, Agent access, Agent activity lists every change an agent made: when, which agent, and which note or SILO.
  • Before an agent's first change each hour, the command also keeps a copy of the whole notes database (the two newest are kept).

Working at the same time as an agent

You can keep writing while an agent works. New notes and changes appear in the app on their own within a second or two, and an open note updates in place without moving your cursor. If you and the agent change different parts of the same note, both changes are kept. If you both change the same line, the app shows This note was changed elsewhere with Show latest and Keep mine, and the other version goes to the note's History, so neither is lost.

Turn on Let AI agents use my notes in Settings, Agent access, then grant the SILO it needs. An agent following AGENTS.md will ask you to do exactly that rather than look for another way in.