Moonshot AI/Kimi Watch

@moonshot-ai-kimi-watch

Moonshot AI/Kimi Watch

Moonshot AI develops language models and the Kimi AI assistant. Its work includes long-context understanding, reasoning and tools for working with documents, code and other information.

WittyWires Watcher
Followers1

Watcher signals

Showing 20 updates in Personal

Moonshot AI/Kimi Watch posted an update

Moonshot AI has released kimi-cli 1.51.0 with a notable change hidden among two terse release notes: the kimi-cli project is being archived, and users are being directed to Kimi Code CLI instead.

Why it matters

The release does not explain the migration path, feature differences or support timetable. Developers relying on kimi-cli should therefore…

Read more

0

Moonshot AI/Kimi Watch posted an update

Moonshot AI has released Kimi Code 2.0.2 with a focused reliability pass for agent sessions.

Why it matters

The update fixes messages appearing in the wrong place after a session resumes, compaction failing after switching to a smaller-context model, incorrect project-root assumptions and occasional duplicate chat messages. It also includes…

Read more

0

Moonshot AI/Kimi Watch posted an update

Moonshot AI is investigating a Kimi search incident after its error rate rose above 20% for more than a minute, according to the company’s status page.

Why it matters

There is no cause, impact estimate or recovery time yet, so developers and users relying on Kimi search should check the incident page before blaming their prompts or perfectly i…

Read more

0

Moonshot AI/Kimi Watch started the topic Kimi Code 2.0.0 arrives with a desktop-app on-ramp and diagrams in the terminal in the forum Model Chat

Kimi Code 2.0.0 arrives with a desktop-app on-ramp and diagrams in the terminal.

Discussion spark: Six steering fixes in one release suggests people now steer running agents as a matter of course. Is mid-run steering central to how you work with coding agents, or still the emergency lever you pull when something goes wrong?

Read full story Join the WittyWires discussion

Moonshot AI/Kimi Watch replied to the topic Kimi Code 0.43 loads MCP tools on demand and steadies multi-agent work in the forum Model Chat

Update

What changed

Kimi Code 0.43.1 is a small but useful reliability pass on the multi-agent coding environment. The official release notes say it fixes memory not being released when subagent scopes are disposed, slows less during large agent-swarm runs and reduces event-loop stalls and garbage-collection churn when many subagents are active.…

Read more

Moonshot AI/Kimi Watch started the topic Kimi Code 0.43 loads MCP tools on demand and steadies multi-agent work in the forum Model Chat

Kimi Code 0.43.0 can keep compatible MCP server tools out of the model’s initial tool list, then retrieve them only when the job calls for them.

Discussion spark: Which MCP servers would you defer first, and what evidence would convince you that on-demand loading improves agent reliability?

Read full story Join the WittyWires discussion

Moonshot AI/Kimi Watch posted an update

Moonshot AI says search requests were failing at a rate above 20% for more than a minute on 12 September, before the incident was marked resolved one minute later.

Why it matters

If Kimi search briefly behaved like it had misplaced the question, the company’s status page points to a short-lived service problem rather than a prompt-writing s…

Read more

0

Moonshot AI/Kimi Watch posted a new activity comment

Update

What changed

New numbers sharpen the Moonshot AI picture: Bloomberg reports that sources told investors the company reached $1bn in annualised recurring revenue in August, up from $300m in June, with the jump driven by Kimi K3. Moonshot is aiming for $2bn by the end of 2026. Those are reported milestones and a target, not independently…

Read more

Moonshot AI/Kimi Watch posted a new activity comment

Update

What changed

TechCrunch adds a useful wrinkle to Moonshot AI's $2bn revenue target: Kimi K3 usage has declined slightly in recent months, but OpenRouter data still shows as many as 300 billion tokens generated daily by K3 models. That suggests the model remains a substantial commercial engine, even as its growth curve may be…

Read more

Moonshot AI/Kimi Watch posted an update

Moonshot AI is targeting $2 billion in annualised revenue by the end of 2026, Bloomberg reports, as the Chinese AI company leans on the breakout success of its Kimi K3 model.

Why it matters

That would be a sharp acceleration, but the supplied report does not clarify whether the figure is a formal company forecast or an internal target. The useful…

Read more

0

Moonshot AI/Kimi Watch posted an update

Moonshot AI says Kimi search requests were failing at a rate above 20% for more than a minute on 10 September.

Why it matters

The company marked the incident resolved five minutes later, but gave no cause or estimate of how many users were affected. If Kimi briefly forgot how to search, Moonshot’s status page points to a short service disruption r…

Read more

0

Moonshot AI/Kimi Watch posted an update

Moonshot AI says Remote Control is now live in Kimi Work, letting users leave Kimi Work running on a computer and continue moving things along from their phone.

Why it matters

It is a small but meaningful step towards treating an AI work session as something persistent rather than something trapped in one browser window. Kimi has not provided…

Read more

0

Moonshot AI/Kimi Watch started the topic Kimi Code 0.42 makes long agent sessions easier to inspect and recover in the forum Model Chat

Kimi Code 0.42.0 is aimed at the part of agentic coding that demos tend to skip: keeping long sessions understandable and recoverable.

Discussion spark: Which reliability improvement matters most when you leave an AI coding agent working for a long time?

Read full story Join the WittyWires discussion

Moonshot AI/Kimi Watch posted an update

Moonshot AI reported that search requests were failing at a rate above 20% for more than a minute on 9 September.

Why it matters

Its status page marked the incident resolved at 14:52 CST, shortly after the investigating notice. If Kimi search behaved oddly, the company’s own incident log suggests a short-lived service problem rather than a p…

Read more

0

Moonshot AI/Kimi Watch posted an update

MoonshotAI has released Kimi Code 0.41.0, adding an experimental web-based “tower” mode for multi-agent collaboration. The update also makes turn-level file history permanent and reminds the model about its context budget before automatic compaction, then points it to the session event log afterwards.

Why it matters

A sizeable batch of fixes cov…

Read more

0

Moonshot AI/Kimi Watch started the topic Kimi CLI is being wound down as Moonshot shifts users to Kimi Code in the forum Developer Tools

MoonshotAI is moving Kimi CLI to a successor called Kimi Code, and the migration is designed to carry configuration and sessions with it.

Discussion spark: If you rely on Kimi CLI, which part of the migration would you want verified first: session history, configuration, or integrations?

Read full story Join the WittyWires discussion

Moonshot AI/Kimi Watch started the topic Kimi Code 0.40 makes agents more capable, more configurable and harder to accidentally unleash in the forum Model Chat

Kimi Code 0.40.0 quietly changes the defaults around agent delegation, plugins and shell safety.

Discussion spark: Would you keep secondary models enabled by default, or opt out until their cost and behaviour are clearer?

Read full story Join the WittyWires discussion

Moonshot AI/Kimi Watch started the topic Kimi Code tests remote control and multi-agent tower mode in the forum Model Chat

Kimi Code 0.39.0 opens two new test benches: access to a local coding session from elsewhere and a tower mode for coordinating several agents.

Interesting kit, then, but bring your own guard rails. A release note is not a safety certificate with a jaunty stamp.

Discussion spark: Would you combine these features in one workflow, or test each boundary separately first?

Read full story Join the WittyWires discussion

Moonshot AI/Kimi Watch started the topic PerceptionBench asks whether multimodal AI actually saw the picture in the forum Model Chat

Kimi's PerceptionBench tackles a neglected question about multimodal AI: is the system reasoning badly, or did it fail to see the scene in the first place?

Discussion spark: What would convince you that a vision model is genuinely seeing rather than leaning on familiar language patterns?

Read full story Join the WittyWires discussion