Profile, memory and skills

Who you are, how you like to work, what to remember and the skills you've written, kept on 0bridge so every AI tool can read them, instead of inside one vendor's app.

Last updated 2026-10-01

Open in Claude Open in ChatGPT

What's in it#

Part	What it is	Limit	How tools get it
Profile	Who you are and how you like to work, as Markdown	4,000 characters	A short excerpt for chat apps; `bridge__profile` for everyone
Global instructions	The instructions you want every tool to follow	32 KB	`bridge__profile`, and your tools' own instruction files
Skills	Folders with a `SKILL.md` and files	100 skills, 256 KB each	`bridge__skills`, and as prompts named `skill_<name>`
Memory	Short facts and preferences, searchable	2,000 items, 2,000 characters each	`bridge__memory_search`, only when asked

Edit it#

0b context profile
0b context push
0b context pull
0b context status
0b memory add "I prefer small pull requests"
0b memory search pull requests
0b memory rm <id>

Your profile is ~/.0bridge/PROFILE.md, and 0b context profile opens it in your editor and then uploads it. The instructions and skills are files in ~/.0bridge too. 0b context push uploads what changed and pull fetches it.
A skill that was never on 0bridge, like one a tool installed for itself, stays on your machine until you choose. 0b status lists the ones waiting. Run 0b skill share <name> to upload one, or 0b skill local <name> to keep it on the machine without being asked again. An interactive 0b context push asks the same question once. Skills you add with 0b skill add, and skills already on 0bridge, sync as before.
A skill marked disable-model-invocation: true is one you start yourself, as in Claude Code. Chat apps aren't offered it on their own; it stays in their prompt menu as skill_<name>, and they read it when you ask for it by name.
If both sides changed, 0bridge keeps your local file and writes the other copy next to it as <file>.0bridge-remote. It never overwrites quietly.
The dashboard's Context page edits the profile and the instructions, shows and removes skills, and lists, searches, adds and forgets memory.
Once you've used 0b context on a machine, the background job (0b background on) keeps that machine in sync. 0b context rm <skill> removes a skill everywhere.

What each client sees#

Chat apps (Claude, ChatGPT): the first 1,200 characters of your profile arrive as the connector's instructions, plus a few lines pointing at memory, skills and resume. Turn the profile excerpt off under Settings → AI apps.
Coding agents already read your instructions file, so they get only the pointer lines.
Memory is never injected. A tool searches it when you refer to past preferences or another AI, and saves to it only when you ask or approve. Chat apps are told to confirm before saving.
Skills are stored, not run. A skill's scripts are files; they run only if your tool chooses to run them.

What 0bridge can read#

Unlike your vault and personal files, the profile, instructions, skills and memory are readable by the server. That is the point: a chat app has to be able to read them. Text shaped like a key is masked on save. If you chose end-to-end history, this still isn't sealed. Keep secrets in the vault, not in your profile.

You can turn memory off for all clients under Settings → AI apps. Deleting your account deletes all of it.

Why not one vendor's memory#

A vendor's memory lives in that vendor's app and follows its rules. This one is plain text you can read, edit and take with you, and it works the same in every tool.

PreviousObjects and skill runs NextConversation history