OpenCovibe

The deciding trait is that every tool call becomes a card with a real diff in it. Instead of hunting through output to work out which file changed and how, you see the change as a change, highlighted, in order, next to the reasoning that produced it. For reviewing an hour of agent work that is the difference between skimming and actually checking.

What you add is a desktop application on top of a command-line tool you already have. Pick it if you review carefully. Pick the plain CLI if you would rather not run one more thing.

A single shared secret is the entire authentication story for remote access, and the thing behind it is not a dashboard, it is control of a developer machine. Tokens leak the ordinary ways: a tunnel URL pasted into a chat, a shell history, a screenshot. There is no second factor described and no per-user identity, so possession of the string is possession of the session.

What it does right is expose permission rules as something you manage rather than something you approve one prompt at a time.

1. Making history addressable turns a conversation into an experiment: the same starting point can be run twice with different instructions, and the difference is attributable to the change rather than to drift. 2. Rewinding the workspace alongside the transcript is the part most implementations omit, and without it a replay restores the words while leaving the files where the failure left them.

2. Nothing is published about what the rewind covers or how far back it holds. The mechanism is documented; its guarantees are not.

Building a marketplace is the correct instinct, because it is the only structure in this product that could ever accumulate switching cost. It is also the hardest thing on the list to start: a catalogue with no buyers attracts no sellers, and a wrapper's catalogue competes with the catalogues the wrapped vendors run themselves.

Moat: the marketplace, if it ever fills. Likely acquirer: none obvious; this is a personal project with good taste. Position: use the interface, do not build a distribution strategy on the catalogue.

Cost analytics per session is the feature I ask every vendor for and rarely get, and having it in a free tool is unusual enough to note. It means the model spend attached to agent work is visible to the person incurring it, which changes behaviour more reliably than a memo from me does.

Packaging is the limit. Two platforms get installers and the third requires compilation, so a fleet rollout covers most desks and not all of them, and there is still no directory login and no audit trail. Approved with conditions: packaged desktops only, remote access disabled by policy.

Switching the model without restarting is a small thing that changes how I work. I can start a session against something expensive, drop to a runtime on my own box for the tedious middle, and go back, all inside one conversation, because the presets are a menu rather than a config file and a relaunch.

Remote hosts over ssh mean the same interface drives a machine in another room, the licence is permissive enough for a fork, and the tool servers I already run are managed here rather than hand-edited into a vendor's configuration.