ThinkRail

The deciding trait is that a workspace is a worktree. Each task gets its own branch and its own directory, and the editor, the changes view and the terminals all follow the one you are in, so switching tasks does not mean re-establishing where you were. The confusion tax of working on two things disappears.

You are the wrong buyer if you only ever have one thing open, because then this is a heavier way to do what your editor already does. Pick it for parallel work. Pick your usual editor for serial.

The agent runs inside the application's own process, and everything a user would complain about, model choice, compaction behaviour, cost accounting and session state, belongs to that upstream project rather than to this one. So a defect in the part people notice is filed somewhere else and fixed on somebody else's schedule, while the crash lands in this window.

What it gets right is scoping sessions. Each tab holds its own agent session, so one confused conversation does not contaminate the next.

1. Making the specification graph read-only separates description from artefact. A view you cannot edit is a view you can trust to reflect something else, and the alternative, an editable graph, would immediately raise the question of which representation is authoritative. 2. Concurrent agent sessions are bound to tabs, so context boundaries follow a visible interface element.

2. No evaluation is published and none is claimed. The design asserts organisation, and the organisation is inspectable in the interface itself.

The vendor is substantial and the framing is the tell. Incubator means the company has ring-fenced this from its committed roadmap, which is prudent for them and information for you: 412 stars is not the number that converts an experiment into a product line.

Moat: the vendor's distribution, if it ever ships properly. Likely path: the ideas graduate into the flagship products and this repository becomes an artefact. Likely acquirer: not applicable; it is already inside one. Position: learn from it, and do not build a team's workflow on an experiment the owner has labelled as one.

A signed desktop installer from an established vendor removes the two objections my endpoint team raises first, which puts this ahead of almost everything else in the category before anyone evaluates the product.

Then it stops. There is no administrative console for sixty machines, no SSO, no audit log and no retention policy, and it does not run unattended, so it produces no delivery metric. Licence cost is zero and the model spend belongs to accounts my engineers hold. Approved with conditions: managed deployment, and no dependence on it in a documented process.

The detail I appreciate is that the command-line tool serves the same application to a browser. That means the desktop bundle is optional, and the thing running on a machine I reach over the network is the same code, which is how I would want to use it anyway.

Apache-2.0 keeps a fork viable. There is no MCP in either direction and no provider configuration of its own, so the model and the tools belong entirely to the agent underneath. This is a shell around somebody else's engine, and it does not pretend otherwise.