Kilo Code on X

Praise comes in superlatives rather than price or benchmarks. Fans call Kilo "the best harness available" for putting several model subscriptions in one place, and one long-time user traces a shared history from Step 3.5 Flash through 3.7 Flash to Step 5 Preview, saying other real Kilo users will know what they mean. A second group values one capability above the rest: notebooks embedded inside Kilo Desktop for research work, permission rules that keep each agent's access separate so nothing posts without approval, and idle-compaction controls a user prefers to a model quietly degrading in a long context.

Free access carries the rest: a weekend experiment pairing a Step 5 preview with the free tier, a week of use described as a great tool that is "currently free through kilocode" with doubt about paying for it later, and a fee-free LLM gateway held up against a competitor's claim. 2 people report still using Kilo or paying for it with no complaint attached, one a paying customer whose team moved over in an afternoon because the export format was just git.

Each of the 3 people here describes a different failure, so nothing in this small set is a pattern. One user who otherwise rates Kilo as amazing and says they love the harness adds that it takes so much memory and freezes sometimes. One asks Kilo directly to fix its login system, complaining that the mail "does not arrive in time, never". One posts the error text from a retry loop, a Novita server overload asking them to try again later, with a trace id — the only one of the three that names a model provider rather than the app itself.

No one in the section says they left over these, names a workaround, or reports the same fault as anyone else. With 3 posts, these are single reports to weigh rather than a list of what breaks for users generally.

With 4 posts and 4 people, one per theme, no comparison in this section represents a group. One says Kilo Desktop is "actually better than OpenCode Desktop" and calls it the best harness for combining all your subscriptions in one place. One prefers VS Code with Kilo Code to a Codex account. One reads a Kilo announcement as a way to stop paying API token pricing and asks whether existing subscriptions can be used instead.

The fourth posts figures for three agents — Kilo at 113k context, 94% cached and 87% tool calls, OpenCode at 91k, 93% and 89%, Claude Code at 89k, 90% and 71% — as that person's own numbers, arguing that such nuances of cache rate and context size matter for both cost and quality. What the section shows is which products Kilo is put next to, not which way people are moving.

Setup talk is the practical half. One user answers how to get xAI Grok running through OAuth: Settings → Providers → Show more providers → xAI, pick xAI Grok OAuth, approve in the browser — no API key is created — then pick a Grok model and test it; for a remote box they add Headless device code and kilo auth login --provider xai. Another asks what the grokbot option does in a particular setting.

The remaining two questions carry no answer in the posts. One asks whether an announcement means they can "avoid API pricing for tokens" by using their existing subscriptions, and one asks whether a 150 requests per day limit is real or a bug on their end. Both are about cost and caps rather than features, and both were put to Kilo itself.

One person asks Kilo directly to "add BYOK support for AssemblyAI". That is the whole of the section — 1 person and 1 post — so it is told as one user's request, not as something customers are asking for. The post is a single line: it asks for bring-your-own-key support for that provider and says nothing about why, or what the user wants to do with it.

Nothing else filed on the page is a feature request. The nearest ask is filed under questions: whether existing subscriptions can be used instead of API token pricing. So the wish list here is one item long, and a reader should treat it as one user's gap, not a ranked demand.