Turn Claude Code, Codex, and Hermes sessions into a shareable graph your agent can query

Coding agents have sped up the process of writing code and made understanding what the agent actually did more time-consuming. Claude Code, Codex CLI, and Hermes Agent all retain full session records on disk, including every tool call, error, retry, subagent, and delegation. This data is too voluminous for a single session, with thousands of lines of JSONL.

To address this, I created rungraph, a free, MIT-licensed tool that scans these transcripts on your disk and presents them in an interactive graph format. The graph visually organizes tool calls in time order, groups them, and assigns subagents to their own lanes. No additional setup or hooks are required, allowing you to immediately analyze runs that encountered issues.

Live sessions also update the graph in real-time. The two main benefits of using rungraph are the ability to query the graph using natural language commands in your terminal and to visualize the answers, making it easier to understand what the agent did and why. The graph also allows you to see the evidence supporting the agent's conclusions, providing a direct link between the claim and the supporting nodes.

This integration of the terminal and the graph creates a seamless workflow where you can ask questions and immediately see the relevant evidence on the visual representation, rather than having to switch between different tools or trust summaries of the run.