Aduzi
A native Windows SSH client that draws the terminal grid on the GPU: one instanced draw call per frame, none at all while nothing changes, and a cold start to a live prompt in under 400 ms. Sessions, keys, SFTP and port forwards, with your own known_hosts and OpenSSH agent, and nothing bolted on beside them.
Install
aduzi.exe --install
aduzi.exe --import
What comes next
From the product's own docs/ROADMAP.md at 216f57a.
The reason to leave the incumbent
- PartialThere is no way to install this client, so it can only be run from a build directory
- PartialThe release archive is built only on this desk, so no signing service can prove it came from the source
- DesignedEvery release still ships unsigned, so SmartScreen warns on the first download and managed machines refuse it
- IdeaIf SignPath Foundation declines the project, there is no affordable certificate lined up
- DesignedOnce a certificate exists, nothing sends the tagged build to SignPath and publishes what comes back
- PartialA search for the name finds the Linux Quickshell and winget's Quick Shell, not this client
- DesignedThere is nothing to download: no release or tag, only a CI artefact that expires in five days
- DesignedA reader who looks the name up finds Portuguese dictionaries and a bean, never why the client is called Aduzi
Recently shipped
All 375 entries →- QS388 · 84e4f47A guest run that starts while the VM is still booting is refused as though the guest had no VMware Tools
Connect-Guest waits up to ten minutes for VMware Tools whether it started the guest or found it still booting, instead of refusing a guest whose Tools say installed.
- QS385 · 9e0cf3eThe description and the site promise a live prompt under 400 ms, while the published measurement is 612 to 647 ms
The GitHub description now states start-up as measured, 647 ms against a 400 ms budget, as the site and llms.txt already did, so no public text claims under 400 ms.
- QS383 · 139d964The shipping discipline tells every task to commit through run-commit.cmd, which the repository no longer has
The skill commits through the PATH run-commit.cmd -m, says it stages the whole tree, and that the tree must hold only the task's files (design superseded: the user chose run-commit.cmd).
- QS378 · c383a4cNobody has checked that every component the archive ships is under an OSI-approved licence
All 18 shipped packages are MIT or ISC, and release.cmd refuses an archive whose notices do not list each at its version under an OSI licence (design recorded in
tools/check-notices.ps1). - QS374 · 5e65ec3The release gate runs while the guest VM a runner left up is still running, and the sitting comes out too noisy to judge
The gate refuses in a second while any VM runs on the desk, and suspend-vm.cmd suspends it once a run holding the guest is done (design recorded in
docs/PERFORMANCE.md). - QS79 · cc0a1b3A change that costs performance is caught by whoever happens to notice it
The perf gate is optional by the maintainer's call: run-perf-gate.cmd or release.cmd -Gated hold parse, emulate and start, and nothing requires it.
- QS376 · 913e9afpublish-release refuses a gated build because the gate's own history row counts as a change to the tree
Both of publish-release's cleanliness checks set benchmarks/results/gate-*.md aside, and the end of a release names the row left to commit.
- QS375 · e11afa9On a quiet sitting the side-by-side gate allows less than the baseline's own noise, and refuses parse 0.7 % worse
A side-by-side figure is allowed the larger of its sitting's spread and the baseline's threshold, while too noisy still reads the sitting alone; parse 0.7 % now holds.