@jakevin7: Any recommendations for currently open-source computer use software, folks? Maka has open-sourced Maka-cu https://githu…
Summary
Maka has open-sourced Maka-cu, a macOS execution layer for AI-driven computer use, forked from an existing project and adapted to support their specific model and interface needs.
View Cached Full Text
Cached at: 09/20/26, 11:20 AM
Any recommendations for currently open-source computer use software, folks? Maka has open-sourced Maka-cu https://github.com/maka-agent/maka-cu… Also know about CUA. Is there anything that’s truly the best practice like that? With implementation results comparable to codex.
maka-agent/maka-cu
Source: https://github.com/maka-agent/maka-cu
maka-cu
The native macOS execution layer for Maka’s Computer Use.
This is a fork of iFurySt/open-codex-computer-use (MIT, © 2026 Leo), an independent reimplementation of the Computer Use service that ships inside Codex. Its accessibility snapshot is the best open one there is; rebuilding that from scratch would have been rework.
Upstream’s own README is kept as UPSTREAM_README.md.
Why a fork rather than a dependency
Upstream is a complete product: a standalone MCP server with its own tool surface, its own permission app, and its own on-screen cursor. Maka already has all three, and has semantics upstream does not:
- an action is bound to the observation it was planned against, is single-use, and a spent one is refused — where upstream re-resolves an element index against a freshly taken snapshot at action time
- the full Anthropic
computer_20251124action contract, because Maka serves models that emit it; upstream is shaped for Codex’s own tool surface - WebContent/renderer elements bound to their real process generation, with unique stale refetch and a trusted background window-event path
So what is taken is the executor: the accessibility snapshot, the dispatch core, input synthesis, the SkyLight background-click path, and app discovery. What is replaced is everything model-facing and everything host-facing, because Maka is already both of those.
Relationship to upstream
Upstream’s effort concentrates on AccessibilitySnapshot.swift. That is the
file this fork keeps closest to upstream and tracks. The parts this fork
rewrites — dispatch binding, permission identity, the tool surface — are ones
upstream rarely touches. That asymmetry is what makes the fork sustainable
rather than a divergence that has to be re-merged forever.
Improvements to the shared accessibility layer belong upstream, not here.
Diagnostics
open-computer-use doctor
open-computer-use doctor --json
Text mode reports protocol, permissions, lock state, native WebContent/SkyLight
capabilities, code signing, and readiness. It opens onboarding when permissions
are missing. --json is side-effect free and emits the same information as one
stable JSON object; add --no-onboarding to text mode for automation.
Attribution
- Forked from
iFurySt/open-codex-computer-useata265277(v0.3.0, 2026-07-27), MIT. - The
sky_clickevent recipe and the private SkyLight bridge originate in trycua/cua (MIT) and yabai. SeeTHIRD_PARTY_NOTICES.md, inherited unchanged. - Behaviour recovered from Codex’s shipped binary is reimplemented, never
copied: that binary is proprietary and confers no license. The archived
copies upstream kept under
docs/references/.../assets/are removed here rather than redistributed.
Similar Articles
@jakevin7: Open-sourced my previous agent: maka, still under intensive development: https://github.com/jackwener/maka-agent… With so many open-source agents already out there, why build another one? - First, maka is a...
Maka is a local-first desktop AI workbench that achieves long-running execution via DAG workflows, with built-in browser automation and plugin capabilities. It is now open-sourced on GitHub.
apache/maka
Maka is a local-first AI agent workspace designed for real work, offering desktop, TUI, CLI, and evaluation features with controlled permissions, runtime logging, and tool integration.
@jakevin7: After joining Apache, Maka has finally successfully released a version as well. The entire community is becoming increa…
Apache Maka, an AI agent workspace incubating at the Apache Software Foundation, has released a version with growing community involvement, featuring local-first design and tool execution capabilities.
@jakevin7: https://maka-agent.com maka 的主页来了!!! 学习了 @wenhaoqi haoqi 老师的网站,用了 @xiaolinbythesea 笑林老师的 SKILL 感谢 @dingyi @Jackywine 两个…
Maka is an open source AI agent that turns user intent into inspectable work, with features like execution, artifacts, tools, and permissions.
@jakevin7: Today, OpenCLI's app was completely rebuilt with a component library, which took a lot of effort. I'm also planning to have the Agent fully refactor the UI of MakeAgent—this might be the final version of the UI. https://github.com/jackwener/maka-…
Maka is a local-first desktop AI workbench built with Electron, supporting multi-model connections, tool calls, permission control, and privacy protection, along with integrated bot access, local memory, and more. The author also mentioned rebuilding the OpenCLI app and planning to refactor MakeAgent's UI.