Juggler

The deciding trait is that you can go back. A session branches into sub-threads, so a promising detour costs you nothing on the main line, and a bad turn is something you back out of rather than something you restart from. Every terminal agent makes you choose between abandoning a run and living with it.

You pay for that in surface area. There is a great deal here to learn before it feels natural. Pick it if you supervise closely and want the controls in front of you. Pick a terminal agent if you would rather type and wait.

The risk is merge semantics deciding what the model sees. A shared document type resolves concurrent edits automatically, by rules designed for prose rather than for a prompt, and nobody reviews the resolution. Two people trimming context at the same time get a result neither of them wrote.

Nothing in the documentation describes conflict handling for that case, and the failure is silent, which is the category of bug that ends up blamed on the model. What it does right is keeping the session on disk rather than in a service, so the record outlives the client that created it.

1. Making the system prompt visible is the choice most products decline, and it changes the epistemics of the whole tool: a surprising answer becomes a question about an input the reader can see. 2. Context is an editable item rather than an accumulated transcript, so the model's view is constructed on purpose.

2. No evaluation is offered and none is claimed, which is consistent, because the argument here concerns control rather than capability. The design should also outlast model changes for the same reason: nothing in it depends on a particular model behaving in a particular way.

There is no business here and that appears to be deliberate. Nothing collects an email, nothing meters a run, and model spend goes to a subscription the user already holds, so the project never sits between a customer and a bill. Admirable, and it means the roadmap is somebody's spare time.

Moat: the extension surface, if enough people write against it before something funded copies the idea. Likely path is a larger vendor shipping the same inspectable session as a feature of a product that already has users. Position: use it, enjoy it, and keep your work portable.

The deployment shape is better than most of this category. A server component runs on a machine we operate with clients attaching to it, so this is a service my platform team can put behind existing controls rather than sixty unmanaged desktops. That is a real difference at my scale.

What the server does not bring is any governance. No accounts, no roles, no retention policy and no record of who attached to which session. Cost is the coding subscriptions we already pay for, which is a number I have. Approved with conditions: the server behind our proxy, one team on it first.

This is the part that matters and almost nobody ships it. The loop strategy itself is an extension, so the thing deciding how the model gets called is code I replace rather than behaviour I live with. Context items and slash commands work the same way, interfaces included.

The licence is the strong copyleft kind, which suits me: whoever ships a modified version owes me the source for it. MCP servers attach, and a local endpoint is a listed provider, so nothing has to leave the machine. There is very little here I could not change if I had to.