deepx-code

The deciding trait is that it starts on the cheap model and escalates only when it has to. Most of what you ask an agent to do is not hard, and paying frontier prices for a file rename is how a monthly bill gets away from you. Here the routing does that arithmetic.

The trade is that you are no longer choosing, and occasionally the cheap model has a go at something it should have handed on. Pick it if cost is the constraint you actually feel. Pick a single-model agent if you would rather be the one deciding what gets the good model.

The precision is language-shaped. Symbol resolution runs through Go's own type checker, which is why callers, interface implementations and impact analysis are trustworthy there. In every other language the row records that the general tools take over, so one feature name covers two very different qualities of answer.

That gap will not announce itself. A user in a TypeScript repository gets an answer from the same command and no signal that it came from a search rather than a type checker. What it does right is naming the mechanism at all, so the boundary is discoverable by anyone who reads.

1. The figure is precise and unaccompanied. A hit rate near ninety-nine per cent requires three declarations to be meaningful: what counts as a long session, what the workload was, and how a hit was counted. None appears. 2. Prefix caching is nonetheless the right architectural bet, because it reduces cost without changing what the model sees.

2. The design is also provider-shaped: the cache belongs to one vendor's API, so the efficiency argument travels only as far as that vendor does. That is a legitimate engineering choice, and it makes the number unportable, which is more useful to know about it than its size.

The dependency is the position. The cost argument, the routing and the defaults are all shaped around one provider's economics, which is a fine engineering decision and a poor strategic one for anybody downstream. If that provider changes its cache behaviour or its prices, the reason to choose this over any other terminal agent evaporates in a release note.

There is no company, no revenue and no funding, so nothing absorbs that shock. Moat: none, unless the provider decides to adopt it, which is the one plausible exit. Position: use it while the arithmetic holds, and keep a second agent configured.

The default is the whole reason I am reading. Native operating-system isolation switched on without anybody choosing it means the risky behaviour is contained on machines where nobody configured anything, which is most machines. Docker is offered as an alternative and so is turning it off.

The rest is the usual shortfall: no single sign-on, no directory sync, no audit export, and it does not run unattended, so it never becomes a pipeline number. Sixty seats cost nothing beyond the provider keys. Approved with conditions: that setting managed centrally, and nobody permitted to turn it off on their own.

Presets for four named model families plus an escape hatch for anything speaking the OpenAI protocol is the configuration I want: the common cases are one line and the unusual case is still possible. My own endpoint counts as a provider here, which means the weights can sit on hardware I own.

MIT and Go means one binary I can build and read, and MCP servers attach so the tools I already run come with me. The workflow scripts follow a convention another tool established, which is unusual generosity in this space. Grudging respect, and no complaints I can turn into a sentence.