introspectintrospectintrospectintrospectintrospectintrospectintrospect

introspect is a window onto how your coding agent is set up and what it’s doing. It reads the files the agent already writes on your machine (system prompt, skills, agents, slash commands, memories, hooks, permissions, plugins) and turns them into a navigable dashboard. You can browse every project the agent has touched, replay a past session as an execution graph, or watch the active one live: a real-time graph of subagent spawns, tool calls, file reads and writes, and current reasoning, streamed over WebSocket.

Built to be safe and legible by design: it is read-only on the agent’s filesystem (it never mutates settings, hooks or memories), binds to localhost only (no network exposure, no auth, no database) and degrades per section, so a missing source shows an empty state instead of crashing. The only thing it produces is an export bundle, to a destination you choose, with secrets redacted by default.