Agent sessions

What is an agent session?

An agent session is one run of an AI coding agent such as Claude Code: what you asked, what the agent read and ran, what it tried and what it found out on the way. It is the working memory of one piece of work.

What a session holds

The conversation, every tool call the agent made and its result, and the decisions it took along the way. Claude Code writes each session as a .jsonl file on the machine where it ran.

Why it gets lost

A session stays on the machine and in the project where it ran. On another machine, for a colleague, or a few weeks later, the next session starts from scratch and works out again what the last one already knew.

Resume a session

Save a session to your a4sx library with a4sx push, then pick up where it left off on any machine with a4sx launch. With no install at all, hand the resume steps to your agent.

Branch a session

Start a new session from any saved one. The saved base is not changed, and the branch nests under it, so you can see which attempt came from where.

Share, buy and sell sessions

From a workspace, publish a saved session to the marketplace for free or with a price. Others read what it contains before they add or buy it, then continue from it in their own agent.

Dead ends count too

A session that never reached its goal still records what was tried and why it did not work. That is worth keeping, and worth sharing while it is current.

• The session library — Save, resume, acquire and publish: how the library works, in detail.
• The a4sx CLI — Install it, save the session you are in, and resume it somewhere else.
• Why keep the dead ends — Failed attempts carry knowledge, and it is worth the most now.
• Questions and answers — Short answers about saving, resuming, buying and selling sessions.