NextClaw

The deciding trait is that the runtime is a choice you make per task rather than a decision you made once at install time. A quick refactor goes to one backend, a long exploration to another, and the conversation, the files and the follow-up work stay in the same place regardless. That keeps the tool from becoming an argument about which agent is best.

What you pay for it is another workspace to learn on top of the CLIs you already know. Pick it if you genuinely switch. Pick your favourite terminal agent if you already know which one you trust.

Abstraction over unlike things leaks in one direction. The backends here differ in tool sets, permission prompts, session semantics and how they report failure, and a single surface drawn over all of them has to either expose the differences or hide them. The documentation names the choice without describing a compatibility matrix, so a user discovers the gaps by switching runtimes mid-project and finding the behaviour changed.

What it does right is keep the task as the durable object. Sources, generated documents and follow-ups survive whichever runtime produced them.

1. Choosing the task as the durable boundary is the decision that separates this from a transcript store, because it keeps an artefact beside the reasoning that produced it and makes the pair retrievable together. 2. Giving each agent its own memory, skills and workspace means context is scoped by role rather than accumulated globally, which bounds what any single run can contaminate.

2. None of this is evaluated, and the interesting measurement would be retrieval quality over a long-lived workspace, which no project in this category publishes.

Subsidised inference with no commercial entity behind it is the least durable arrangement in this category. A trial gateway is a genuine kindness to new users and it is also a bill that grows with adoption, which means the more successful the project becomes the sooner that door closes. Nothing in the repository describes who absorbs it or for how long.

Moat: none; the runtimes underneath belong to other companies. Likely path: the gateway is withdrawn and the project becomes key-only. Position: connect your own provider on day one and treat the trial as a demonstration.

A tool that can be invoked non-interactively with machine-readable output is a tool I can put in a pipeline and measure, rather than a desktop habit I can only survey people about. That is genuinely rare on this board and it changes what the thing is for: scheduled checks, batch migrations, work that produces an artefact somebody reviews.

Identity is still absent. No single sign-on, no provisioning, no audit trail and no console, so sixty installations are sixty configurations. Approved with conditions: it runs as a service account in our pipeline, and desktop use stays individual and optional.

Naming a self-hosted serving stack among the supported providers rather than burying it in an advanced section is the signal I look for. It means somebody ran it that way, and it means the workspace, the scheduling and the browser control all work with the weights sitting on hardware I own.

Deployment is my choice too: a package on my machine, a container, or a box I rent. Permissive licence, protocol client included, and the applications it builds stay on disk afterwards instead of evaporating with the session.