Aide

This one is over, so the recommendation is short. There is nothing to try daily, no updates arriving, and no reason to move an editing setup onto a build that will drift further from the world every month. If you downloaded it in 2024 and liked it, that memory is the whole of the value now.

Pick Void if you want an open editor with an agent inside it and a project still shipping. Pick Cursor if you want the polished version of the idea Aide was chasing and can live with a subscription.

The failure mode is not a bug, it is the archive banner. An archived repository accepts no patches, so every defect present at the moment it froze is permanent. A downstream fork of a large editor inherits its parent's fixes only while somebody merges them, and nobody is merging them. Time makes this worse on a schedule.

Treat any install as frozen software and plan a migration rather than a workaround. What it does right: building on Code - OSS meant the extension ecosystem worked, so the thing users depended on was never proprietary to begin with.

The benchmark claim is the only quantitative statement attached to this project, and it is unauditable as recorded. Three things a reader would need are absent: which model produced the run, what scaffold surrounded it, and how many attempts each instance received. Without them a verified-subset number cannot be placed next to any other number on this board.

The documentation is now thin in the strict sense, since the repository is read-only and the site describes a research programme rather than the editor. The honest summary is that a result was announced and cannot be checked.

The commercial design was thin from the start: give away an editor, charge for the inference behind it, and hope the margin between wholesale and retail tokens covers a team. That spread belongs to the model vendors, and every company that tried to live in it discovered the same thing at roughly the same time.

What remains is a research identity, which is what a good team does while it looks for the next thing or the next buyer. Likely outcome: an acqui-hire into a lab that wants the people. Position: no exposure, and watch where the founders land.

The procurement answer writes itself. There is no supplier to contract with, no single sign-on, no audit logging, no data-retention statement and nobody to escalate to when something breaks at eleven at night. Deploying an unmaintained editor to sixty machines means owning every future vulnerability in it ourselves, forever, with no budget line attached.

If anyone here already runs it, that is a removal ticket, not a rollout. Onboarding cost is irrelevant when the exit cost is the only number that matters. Not yet, and not later.

The licence is the best thing here. AGPL-3.0 means what I have is the whole editor and a fork stays open, which is more than most abandoned products leave behind. If somebody wanted to restart this, the legal path is clear.

Everything else disappoints. No MCP on either side, so my tools do not reach it. No local model support, so the interesting half of my setup is unusable. Bringing a key was possible, but the good version of the product ran on their servers, which is exactly the arrangement I avoid. A fork could survive the vendor. Nobody has started one.