better-codex

The problem it addresses

Better Codex is a native task board and multi agent collaboration system that runs inside the Codex desktop app. It is local first and open source, and it targets macOS and Windows with a one command install. The author frames it as a work layer on top of Codex, because the model itself is already strong but the surrounding workflow is thin. The README lays out the pain points that motivated the tool. First, the session list becomes a graveyard of near identical conversations, none grouped by the actual project, so finding a past thread means scrolling and guessing. Second, one global model configuration rules every new session, so tuning the reasoning level for a hard refactor then slows down quick question sessions. Third, ideas that appear mid conversation have nowhere to live, and they end up in a notes app or disappear. Fourth, Codex is shaped like a chat log, but real work is a project that spans days and many threads. None of these issues is the model's fault, the README argues. They are about organizing the work around the model. Better Codex answers them by giving conversations a structure: tasks, projects, owners, and a board that shows what needs attention. This turns a flat list of chats into a tracked body of work where each item links back to the conversation that produced it.

What you get

The feature set maps directly to the problems above. You can turn conversations into tasks and organize them by project, status, and owner, then open a task to return to its linked conversation. Ideas can be captured as new tasks the moment they arrive, so they are not lost between sessions. Task progress is visible at a glance: you assign work to yourself or to an Agent, run it manually or automatically, and tasks return to the board when they need review, a decision, or unblocking. Each Agent can be configured on its own. The README lists the settings per agent as the model, reasoning level, permissions, instructions, and concurrency limit. That means a research agent and a drafting agent can run with different models and different guardrails inside the same workspace. The README describes who benefits. Developers manage requirements, bugs, refactors, and release checks by project. Solo companies keep product, customer, and content work on one board while different agents handle research, drafting, and checking. Content creators save ideas and track outlines, drafts, and revisions. Product managers, sales people, and office staff get similar structure for their own recurring work. A short walkthrough in the README shows opening the Task board from the Codex sidebar, creating a project, and capturing a task, either manually or straight from a conversation.

Install and supported platforms

Better Codex installs with a single command, which the README presents as one of its main conveniences. Because it runs inside Codex Desktop, the board and the agent configuration live where the conversations already happen, rather than in a separate app you have to switch to. The supported platforms are macOS, covering both Apple Silicon and Intel, and Windows on x64. The README badge set shows release, star, download, and build status, plus the two platform marks, so a new user can confirm support before installing. The project also ships a Chinese language README, which suggests the author expects a broader audience than English only users. Being local first means the work stays on your machine rather than syncing to a vendor cloud by default, which matters for people handling private or client code inside Codex. The agent model is the same Codex you already use, just wrapped in reusable profiles and a board, so there is little new to learn about the underlying model. The combination of a one command install, local storage, and per agent configuration makes it a low friction add on for anyone who already spends their day in Codex and wants their sessions to add up to tracked progress.

Editorial conclusion

The project is released under the Apache-2.0 license and is written in TypeScript.