rungraph
fayzan123/rungraph
Turns local Claude Code, Codex, Hermes, opencode and Cursor session logs into an interactive graph
rungraph is a local command-line tool that reads agent session transcripts, including Hermes Agent runs, and draws each run as an interactive graph of turns, subagents and tool calls. It helps Hermes Agent users see where a run retried, failed or needed human input.
What rungraph does
Running npx rungraph finds Claude Code, Codex, Hermes Agent, opencode and Cursor runs still on disk, starts a local server and opens the browser. There are no hooks, wrappers or telemetry, so it works on past runs, and a run in progress grows live. Prompts form the backbone of the graph, parallel agents fan out into lanes, and tool calls appear as nodes such as Bash and Edit, with consecutive calls of the same tool collapsed into one node. Clicking a node shows its inputs, outputs, errors and timing.
rungraph derives signals from each run: retry storm, unresolved error, human intervention, outlier cost and course change. A coverage figure reports how many records it could interpret. Through an MCP server installed with npx rungraph mcp --install, the agent can query the graph itself, and the nodes an answer refers to light up in the open view. Hermes, opencode and Cursor runs need Node 22.13 or newer for the built-in SQLite reader.
Key features
- One command, npx rungraph, scans local session data and opens a viewer
- Nodes for turns, subagents, tool calls and human interventions
- Signal strip for retry storms, unresolved errors, interventions, outliers and course changes
- Coverage figure showing how many transcript records were understood
- MCP server so the agent can query the graph and highlight nodes
- Find box and file view to trace which steps touched a path
When to use it
- Work out why a tool call kept failing in a long Hermes run
- Review where you denied a permission or interrupted a run
- Inspect how subagents were spawned and what they returned
Who it is for: Developers who run Hermes Agent, Claude Code, Codex, opencode or Cursor and want to inspect what a session actually did.
How it fits with Hermes Agent
A general tool that reads Hermes Agent runs alongside transcripts from several other coding agents; it is not built only for Hermes.
How to install rungraph
These commands are copied from the project's README. Check the repository for the latest steps before you run them.
npx rungraph
npx rungraph mcp --install
npx rungraph mcp --checkRequirements: Node 22.13 or newer for Hermes, opencode and Cursor runs; on older Node versions those runs are skipped with a warning.
FAQ
What is rungraph?
rungraph is a local tool that turns agent session transcripts into an interactive graph. It shows prompts, subagents, tool calls and the moments a human said no, and it flags retries and unresolved errors.
Does rungraph work with Hermes Agent?
Yes. It finds Hermes Agent runs on disk along with Claude Code, Codex, opencode and Cursor runs. Hermes runs need Node 22.13 or newer.
Is rungraph free and open source?
Yes. The repository is published under the MIT license.
Similar dev tools for Hermes Agent
All dev toolsmacOS menu bar app that lights up your screen edges while Claude Code or Hermes Agent is working
magicwubiao go-magicGo AI agent framework inspired by Hermes Agent, with a TUI, web dashboard and messaging gateway
freu-ai Freu CLIRecord a browser workflow once and compile it into a reusable skill for Hermes and other agents
bpauli gccliGarmin Connect command-line client with JSON output and a bundled agent skill for Hermes
GodsBoy Jev Agent Skill RouterTyped, confidence-aware routing that picks which skill an agent should use, or none
AlekseiUL Agent Loop Engineering KitDesign, validate, dry-run and audit bounded Hermes Agent loop contracts before automating them
Related guides: How to install Hermes Agent · Run multiple Hermes agents with profiles