Manage shared Memory files

Memory is a shared UTF-8 filesystem for your workspace. People and coding agents can create, organize, search, read, and edit text files without copying them into a local project checkout. Open Settings → Memory to browse the whole filesystem. Open a Project’s settings to work inside that Project’s folder.

Understand the filesystem

/
├── workspace/
│   ├── instructions.md
│   ├── working-memory.md
│   ├── integrations/
│   └── skills/
├── projects/<project>/
│   ├── instructions.md
│   ├── integrations/<provider>.md
│   ├── skills/
│   └── other files you create
├── archived/<project>/             read-only archived Project files
└── legacy/                         read-only imported and retired content
    ├── workspace/
    └── projects/<project>/

Location	Purpose	Who can change it
`workspace/`	Files shared across the workspace.	Project Runs, reception Runs, and members with workspace content write access.
`projects/<project>/`	Instructions, integration notes, skills, and free-form files for one Project.	That Project’s agents and members with workspace content write access.
`archived/<project>/`	The Project folder and files preserved when a Project is archived. Unarchiving moves the same folder back to `projects/`.	Nobody while the Project is archived; it is read-only.
`legacy/`	Exact snapshots imported from former settings fields, plus complete folders retained when a Project is permanently deleted.	Nobody; it is read-only.

Connected workspace providers get an empty integrations/<provider>.md file in each Project when no previous instruction exists. Disconnecting an Integration does not delete that file or its edits. Archiving a Project moves its folder to archived/; unarchiving returns it to projects/. Permanently deleting a Project removes it from the active and archived Project lists, but keeps its folder and files read-only under legacy/projects/.

Cogno reserves only the locations shown above. Folders or files such as notes/, agents/, and README.md are free-form content created by a person or agent, or preserved from an earlier trial. Cogno does not create them as part of the standard layout.

Memory files are not automatically inserted into every prompt. Coding agents list, search, and read the files they need. A Run with a Project reads its own and other public Project folders in the same Workspace, plus shared /workspace. It edits its own Project folder and shared /workspace Memory, including Skills. Other locked Projects stay hidden even if the person running it is a member or admin. A locked Project’s own Run can still read public Projects. Relative paths begin in its Project. Project-less reception Runs retain their existing Workspace scope. An agent can browse folder names in small pages before opening the files it needs, even when Memory is too large to load into one shell command.

Edit Memory files

Open Settings → Memory

Open Settings → Memory. The Explorer shows workspace, live projects, archived Projects, and legacy in that order. The archived and legacy sections start collapsed.

Create or open a file

Select New file or New folder, or select an existing file. Markdown files open in Preview or Markdown mode; other UTF-8 files open as Plain text.

Edit and let Cogno save

Type in the editor. Changes save automatically after you stop typing; ⌘S saves immediately.

Copy the agent-visible path

Select Copy path when you want to reference the exact path in a task or conversation.

Memory checks for updates while you keep the page open. Unchanged files reuse the version already loaded, and only the selected file is opened. Your current edit stays in the editor when a newer version is detected. If an agent or another member saves the same file while you edit it, Cogno shows This file changed while you were editing. Choose Load latest or Overwrite with mine. Cogno never silently overwrites the other version.

Deleting a file or folder cannot be undone. Deleting a folder also deletes every file inside it.

Import instruction files during desktop setup

When you select local folders during desktop setup, Cogno looks for their top-level CLAUDE.md and AGENTS.md, plus your personal ~/.claude/CLAUDE.md. If it finds any, Add CLAUDE.md / AGENTS.md to memory offers to copy them to the new Projects. The option starts off.

Imported files become shared Project Memory: workspace members and agents can read them, including your personal instructions. Review the source files first. Leave the option off if they contain private information or secrets.

Folder instructions are saved under imported/folders/<folder>/; personal instructions use imported/personal/CLAUDE.md. Existing files are never overwritten. Empty, binary, or oversized files are skipped. Existing agent Projects, conversations, and other agent memory are not imported. If a copy fails, the Project remains created and the setup screen shows the failure. Select Retry before leaving the screen to retry only the failed files without creating another Project. A name conflict remains a failure and does not overwrite the existing file. The retry list is kept only while this setup screen is open.

Teach agents with skills

A skill is a folder with a SKILL.md file that tells coding agents how to do one kind of work. Put it in Memory and every agent run for that workspace or Project gets it in Claude Code’s or Codex’s own skill list, next to the skills already on that computer. The agent first sees a skill’s name and description, then reads its full instructions when it selects it. Cogno’s built-in chat-reply covers chat requests, summaries, updates and feedback about length or presentation. You do not need to create another skill for each conversation. Cogno replies in chat in plain words, in about five lines. For a list of more than three items it gives the top three and puts the rest in a Task. This applies in Slack, Microsoft Teams, Discord and Google Chat. When you say how you want replies to look or sound, Cogno keeps it as one short line for that person or place in Memory. It edits that line when you say something new instead of adding a dated note, and when someone mentions Cogno it reads the line before its first post. Event-triggered skills. A skill with an event entry in its .cogno.yaml runs only when a message is an actual instance of its when condition. Nearby talk, talk about the situation, and a message that withdraws or corrects an earlier one do not trigger it. When Cogno creates such a skill, it keeps when to one specific situation and uses a channel where Cogno can post. Cogno also keeps a skill’s own posted output to 1 to 3 lines.

Where you save it	Which runs get it	Listed as
`workspace/skills/<name>/SKILL.md`	Every run in the workspace	`<name>`
`projects/<project>/skills/<name>/SKILL.md`	That Project’s runs	`<Project>:<name>`

The skills folder does not have to sit at the top. Any folder named skills counts, at any depth: workspace/teams/sales/skills/proposal/SKILL.md is the workspace skill proposal, and projects/acme/docs/skills/release/SKILL.md is a skill of the Project acme. The top folder decides who gets it: everything under workspace/ is for the whole workspace, everything under projects/<project>/ is for that Project. If two skills in the same place have the same name, only the first loads. Agents can save their own skills under workspace/.cogno/skills/ or a Project’s .cogno/skills/. Expand .cogno and then skills in the Memory Explorer to open and edit those saved files. The .cogno parent is available for navigation; it cannot be renamed or deleted. Other hidden folders (names starting with .) stay hidden from people; skills inside them still load for agents. Cogno’s built-in skills are not listed in the Memory Explorer. A new skill under workspace/.cogno/skills/ needs a name no built-in uses. Write the file in this shape:

---
description: When to use this skill, in one or two sentences.
---

# Steps

…

Rule	Detail
`description`	Required, up to 1,024 characters. It is what the agent sees when it chooses a skill.
Name	The folder name, or a `name:` line in the header. Lowercase letters, digits, and single hyphens; up to 55 characters; may not contain `claude` or `anthropic`.
Other files	Files beside `SKILL.md` (for example `scripts/`) come along with the skill.
Changes	A saved change reaches running agents on their next turn.

When a SKILL.md would not load, the editor shows Saved, but agents won’t load this skill yet. with what to fix. The file is still saved; fix it and save again. Skills already on the computer are never hidden. When a name is taken, Cogno’s skill gets a suffix instead:

Situation	Name in the agent’s list
The computer already has a skill or command with that name	`<name>-cogno`
A workspace skill has the same name as one of Cogno’s own	`<name>-ws` (or `<name>-cogno-ws` when both apply)
A Project’s name contains spaces, `:`, `/`, or `@`	Those become `_`: `Sales Ops:<name>` is listed as `Sales_Ops:<name>`
You have a plugin with the Project’s name	`<Project>-cogno:<name>`

A Scheduled Skill is a skill like any other: a skills/<name>/SKILL.md folder inside a Project or under /workspace/skills/, with a .cogno.yaml beside it that has a schedule entry (a cron, for example - schedule: "0 9 * * 1"). It is handed to agents like any other skill and also runs on its schedule. The entry can also carry timezone and enabled; a Workspace Skill can name an execution Project with project. Agents can check the schedule’s status, next and last run with list_scheduled_skills. A Project run checks its Project and shared Workspace Skills; a reception run with no Project checks the Workspace. A one-time run is not a Scheduled Skill. A flat skills/<name>.md file is not a skill.

Describe the routine you want to change or stop in a conversation Cogno receives. You do not need its file path or the original thread. If several routines match, identify the intended destination or schedule. Cogno’s built-in broad checks for important work and repeated manual work each have a daily baseline schedule. Their nominal time is 09:00 in the Workspace timezone, with checks spread across Workspaces rather than all starting together. Specific deadlines, pending Tasks and promised replies can get an earlier, focused check. The important-work survey ends without a no-change post when it finds no useful change or unresolved work. Your own schedule overrides and stopped routines stay in effect.

Reference

Property	Behavior
Content	UTF-8 text files only. Binary files are not supported.
Paths	Case-sensitive. `.` and `..` path segments are not accepted.
Agent reads	Workspace scope reads the Run’s own and public Project folders plus `/workspace`. Project scope loads only its own folder. Other locked Projects stay hidden regardless of human access.
Agent writes	`memory_write`, `memory_edit`, and filesystem edits can change the Run’s own Project folder and shared `/workspace` Memory, including Skills. Other Project folders remain read-only. Project-less reception Runs retain their existing Workspace scope.
Project visibility	Project Settings shows that Project’s subtree; Workspace Memory shows the entire filesystem.
Saving	Autosave with conflict detection based on the file version that was opened.
Migration	Existing instructions and working memory are copied verbatim; Cogno does not summarize or split them. Orphan folders from previously deleted Projects are preserved under `legacy/projects/`.

Start

Release notes

Concepts

Guides

Reference

Help

Manage shared Memory files

Understand the filesystem

Edit Memory files

Import instruction files during desktop setup

Teach agents with skills

Reference

​Understand the filesystem

​Edit Memory files

​Import instruction files during desktop setup

​Teach agents with skills

​Reference

​Related docs

Understand the filesystem

Edit Memory files

Import instruction files during desktop setup

Teach agents with skills

Reference

Related docs