Price terms as published

Each line is the vendor’s own wording, split from the price capture checked on 2026-10-14. Nothing here is rounded, converted or re-worded — the full sentences stay in the review below. Lines that carry only a price continue the plan named on the line above, in the vendor’s own order.

2 price terms for Mano-P
#Plan / termChecked
1Free, open-source (Apache-2.0)2026-10-14
2cloud API pricing not publicly listed2026-10-14

One-liner: Mano-P is a computer-use agent (CUA) from Mininglamp Technology (HKEX: 2718) that operates GUIs by looking at screenshots and moving the mouse — like OpenAI’s Operator, but a small specialized model you can run locally, under Apache-2.0.

Verified facts (checked 2026-09-07; repo stats re-read 2026-10-14)

Attribute Value Source
GitHub stars / forks 2,821 / 274 GitHub API, 2026-10-14
License Apache-2.0 GitHub API, same day
Repo created 2026-03-25 GitHub API
Last push 2026-06-25 GitHub API, same day
OSWorld, specialized-model track #1 — 58.2%, 13.2pp ahead of #2 (opencua-72b at 45.0%) official README / vendor announcement
OSWorld, all models 5th — Claude Sonnet 4.6 leads at 72.1% README FAQ (vendor’s own admission)
Deployment forms CLI, Python SDK, Agent Skill (mountable in OpenClaw / Claude Code) GitHub README
Cloud endpoint mano.mininglamp.com (root returns FastAPI JSON 404 — service up, no public landing page) probed 2026-09-06

Three things the hype posts got wrong

  1. “13 leaderboards SOTA” — true only with the qualifier the vendor itself applies: among small/specialized models. The README’s own FAQ concedes general frontier models (Claude at 72.1%) beat Mano on OSWorld overall by a wide margin. Any benchmark citation should carry that qualifier.
  2. “Open-sourced less than a month ago” (Douyin video claim, 09-06) — the repo was created 2026-03-25 and last pushed 2026-06-25. By the video date the release was over five months old and the repo had been quiet for two+ months. GitHub API doesn’t lie.
  3. “Your data never leaves the device” — only true after you manually switch off the default. The README states plainly that cloud mode is the default, and cloud mode means your screen screenshots are uploaded to mano.mininglamp.com. Local inference has hard hardware requirements: an Apple Silicon M4-class Mac with 32GB RAM (or a USB4.0 compute stick); Windows/Linux support is labeled beta by the vendor.

Cloud API pricing

There is no public pricing or signup page for the hosted API as of 2026-09-07 — the cloud root URL returns an unregistered-route JSON 404. Whatever the marketing videos imply about “just call the API,” the commercial terms are not discoverable from outside. This entry gives no rating to the cloud offering for that reason.

Who it’s for

  • Researchers/tinkerers who want a competent, permissively-licensed GUI agent they can actually read the weights story of
  • Teams on M4 Macs wanting on-device automation without frontier-model API costs
  • Not (yet) for: anyone needing Windows/Linux at production quality, or anyone who assumed “local-first” described the default experience

Sources