Codelux AI next to Claude Code, Cursor, Copilot and Codex
These are coding agents. Codelux AI is not one: it is the space they work in together — one memory and one history of the project for every agent, whichever vendor it comes from, and no agent overwriting another's work. It is meant to run alongside one or more of the tools below.
Facts as of October 2026, from each vendor's documentationWhat each one offers for work across agents and sessions
A dash means the vendor's documentation does not cover it as of October 2026 — not that the product cannot do it.
| Codelux AI | Claude Code | Cursor | GitHub Copilot | OpenAI Codex | |
|---|---|---|---|---|---|
| What it is | An app on your computer, next to your AI tools; a plugin for Claude Code | Anthropic's coding agent, in the terminal and IDEs | An AI code editor with agents, local and in the cloud | GitHub's assistant, with a cloud agent that works on issues | OpenAI's coding agent, as a CLI and cloud tasks |
| Project knowledge across sessions | A memory each agent writes at the end of a task (what, why, next step, traps, facts at file:line), read back by every agent on the project |
CLAUDE.md files and an “auto memory” of notes Claude writes, loaded at session start |
Project, user and team rules, and AGENTS.md |
Custom instructions, Spaces, and Copilot Memory (public preview) for repository facts | AGENTS.md; memories are off by default and stored locally |
| Several agents at once | AI tools of different vendors on the same project, without overwriting each other | Subagents with their own context; isolated git worktrees; agent teams (experimental) | Parallel agents, each in its own git worktree; cloud agents in isolated machines | The cloud agent runs each task in its own environment and opens one pull request per task | Cloud tasks, also started from Slack or Linear |
| Two agents on the same file | One agent edits a file at a time; the others are told who has it. Across colleagues' computers on Team | Isolation by worktree; agent teams lock task claims, and the docs advise giving each teammate different files | Isolation by worktree: the docs say agents do not share files | Isolation by branch and pull request | — |
| History of what agents did | One timeline of memories, notes, edits, commits, locks and messages, with person and agent; export as JSON or CSV | Optional OpenTelemetry export of events | — | Session logs of every step of the cloud agent | — |
| Where it runs | On your machine; team data end-to-end encrypted; optional company server | On your machine; cloud sessions available | Desktop app; cloud agents | IDE extensions and GitHub; the cloud agent runs on GitHub Actions | CLI on your machine; cloud tasks |
How Codelux AI fits with each of them
With Claude Code
The plugin loads the project memory at every session start, protects Claude Code's edits from other agents and reminds it to save what it did. Your CLAUDE.md keeps working as before.
With Cursor
One line of configuration. Cursor's agent reads and writes the same memory, notes and history as the other agents on the project.
With Codex and Gemini CLI
Ready-made configurations give Codex and Gemini CLI the same automatic memory as Claude Code (experimental).
With other editors
VS Code, Windsurf, JetBrains, Zed and any tool that supports MCP use the same memory, notes and history.
When you need it, and when you do not
It helps when…
- two or more agents — or two people's agents — work on the same project
- you switch between agents and want the next one to start where the last one stopped
- you want one record of what each agent did, and why
- you run several projects and want the same memory and history on each
You may not need it when…
- you use one agent, in one session at a time, and its own rules files are enough
- each agent always works in its own worktree or branch and never on shared files
Try it next to the agents you already use
Free on one project, every feature, no time limit, no card. For organizations, see the team comparison.