Cogpit does not flatten Claude Code and Codex into a lowest-common-denominator chat box. It gives both providers a consistent operator experience while preserving the controls each runtime does best.
Choose a provider below to inspect the native control surface Cogpit connects to.
Stream turns, tools, reasoning, and agent activity directly from Claude's runtime.
Follow native goal progress, task owners, dependencies, elapsed time, and completion state.
Rewind files to a Claude checkpoint, with deterministic edit reversal as a fallback.
Inspect background agents, open their sessions, and stop active tasks from the dashboard.
The interface stays familiar even when the provider underneath changes.
Create new work or return to sessions already on disk.
Reasoning, assistant text, tool calls, results, edits, and sub-agent activity.
Add direction to active work or stop the current turn without killing the session.
Review provider-native command, file, network, and tool requests in context.
Use the models, reasoning levels, speed tiers, and permission modes the runtime advertises.
Reopen history with its timeline, usage, file changes, and agent relationships intact.
Cogpit connects through the Claude Agent SDK for active work and reads Claude's session history for durable restoration.
Cogpit uses Codex's persistent app server for threads, turns, steering, approvals, and capability discovery.
Cogpit asks your installed CLI what it supports. Controls appear only when the selected model and account advertise them, so new reasoning levels, image support, and speed tiers can arrive without turning the interface into a compatibility minefield.
Get Cogpit