Session Atlas turns a Claude Code session into an interactive map of what the model did: every prompt, every tool call, every shell command, and why. It also shows usage across all your sessions and flags risky actions, such as a force push or a read of your .env file.
These docs are for people who use Session Atlas. They do not cover installing or operating the server.
Start here#
| I want to… | Read |
|---|---|
| Run the server | Setting up the server |
| Get Session Atlas running and see my first session | Getting started |
| Send sessions to the server, one or all | Pushing sessions |
| Understand a session and follow what the model did | Exploring a session |
| Search every shell command the model ran | Searching commands |
| See usage and trends across sessions | Dashboard |
| Get warned about risky actions | Rules and alerts |
| Fix a problem or check privacy | Troubleshooting and privacy |
How it fits together#
Claude Code ──/push-trace──▶ Session Atlas server ◀── your browser
- You work with Claude Code as usual.
- You run
/push-traceto upload a session’s transcript to the server. - You open the server’s web address in a browser to explore it.
Words used in these docs#
| Term | Meaning |
|---|---|
| Session | One Claude Code conversation, from start to end. |
| Prompt / turn | One message from you and everything the model did in answer. |
| Tool call | One action the model took: running a command, reading or editing a file, searching, and so on. |
| Subagent | A helper the model started for part of the work. It has its own tool calls. |
| Context | How much text the model was given for a request. It grows as the session goes on. |
| Rule | A saved check that flags matching tool calls. |
| Alert | A tool call that a rule matched. |