Resume anywhere

Continue a coding session in a different AI tool, on a different machine, or from a chat app: copy one short reference, and the next tool picks up the goal, the open items and the last turns. This needs conversation history turned on: Conversation history.

Last updated 2026-10-01

The reference#

Every uploaded session gets a short reference like 0b:k3f9x2. The dashboard's History page shows it with a copy icon. The icon copies continue 0b:k3f9x2, which any AI connected to 0bridge understands: paste it into a chat or a terminal agent and it calls bridge__resume.

0bridge accepts several spellings of the same session:

• the short reference, 0b:k3f9x2, or just k3f9x2;
• the full id, such as claude-code:<uuid>;
• the start of the tool's own session id, from 8 characters.

If more than one session matches, you get the candidates to pick from.

From a terminal#

0b resume 0b:k3f9x2

• On the machine where the session ran, and with the tool that ran it, 0b resume offers the tool's own resume (claude --resume, codex resume) in the original folder.
• Anywhere else it writes the handoff to ~/.0bridge/handoffs/ and starts the tool you choose with it as the first message: --tool claude, codex, gemini or cursor.
• --print shows the handoff instead of starting anything. Inside an agent, or when the output isn't a terminal, that is the default.
• --turns n sets how many recent turns it includes.

From an AI tool#

In a coding agent, or in a chat app that has chat history turned on, say continue 0b:k3f9x2. The tool calls bridge__resume and gets the handoff as text.

To leave a good handoff, ask the tool to save a checkpoint before you switch: "save a checkpoint of this session". It calls bridge__session_save with a summary and the open items, and they head the handoff. A chat app can also save a short summary of a conversation that never touched your machine, and you can continue that anywhere.

What the handoff contains#

It is built without any AI model on 0bridge's side, so it is the same every time:

• the session's title, tool, repo, branch, machine and date;
• the summary and open items, if a model saved them;
• the goal: your first message, up to 600 characters;
• the last 6 turns, each at most 1,500 characters and 8,000 in all.

It tells the next tool to check the repo's current state before acting and not to redo finished work, and names the call that fetches the full transcript.

What it can't do#

• It doesn't move a live process. The next tool starts a new session that knows what the old one did.
• It doesn't carry tool calls or output, only the conversation, as the history upload does.
• Sessions in end-to-end mode are sealed: a chat app can't open them. Use 0b resume on a machine that has your vault key.
• Chat apps need Chat history turned on under Settings → AI apps, and a project that limits history to its own repo applies that limit here too.