Kaeso

@logicplanes

[Kaeso @kaesoapp]

Fast models deserve a fast gateway. 0.5.6 also translates streamed replies 1.3 to 1.6x faster and decodes large requests about a third faster.

https://t.co/Aj1AoN1xtd

[Kaeso @kaesoapp]

Cerebras is now a built-in provider in Kaeso.

Add your key and call Cerebras models from any app format: Chat Completions, Responses or Messages.

Out now in v0.5.6.

[Kaeso @kaesoapp]

@dvnold @superlogical @Spotify @SlackHQ clean

[Devin Oldenburg @dvnold]

asthetic maxing with @kaesoapp

[Kaeso @kaesoapp]

Shipped Kaeso v0.5.0

Ask twice, pay once: projects can now turn on a response cache. A repeated identical request gets the stored reply, at no provider cost.

[Kaeso @kaesoapp]

@marchoeffl Usage, limits and budgets don’t depend on ClickHouse. You only lose the per-request detail, not the spend tracking.

[Kaeso @kaesoapp]

Shipped Kaeso v0.5.0

Ask twice, pay once: projects can now turn on a response cache. A repeated identical request gets the stored reply, at no provider cost.

[Kaeso @kaesoapp]

@chloekmkyc @logicplanes Thanks!

[Kaeso @kaesoapp]

Model data now comes from KPMR, our own registry, as signed catalogues instead of https://t.co/fu4aOWWilC. A catalogue no trusted key verifies is refused.

Upgrading: rename models_dev settings to kpmr. The old ones still work in 0.5.

https://t.co/ZPdcAUKgZI

[Kaeso @kaesoapp]

ClickHouse is now optional. Run Kaeso without request logs, or point it at your own ClickHouse.

Without it, the gateway needs about 1.8 GB less memory under load. Usage, limits and budgets work the same.

[Kaeso @kaesoapp]

Your app speaks Chat Completions, Responses or Messages. Kaeso translates, so Kimi K3, Qwen3 Max, Command A Plus or GLM-5.3 work with the SDK you already use.

QwenCloud brings its embeddings too.

[Kaeso @kaesoapp]

Shipped Kaeso v0.4.8

Four new built-in providers: Moonshot (Kimi), QwenCloud, Cohere and Ollama Cloud.

Add a key and call their models from any app format.

[Kaeso @kaesoapp]

Also in 0.4.8: smoother dialogs, and custom providers and SSO discovery can never reach Alibaba Cloud's metadata service.

Full changelog: https://t.co/hLOL1S1Sk5

[Kaeso @kaesoapp]

Also in 0.4.8: smoother dialogs, and custom providers and SSO discovery can never reach Alibaba Cloud's metadata service.

Full changelog: https://t.co/hLOL1S1Sk5

[Kaeso @kaesoapp]

Shipped Kaeso v0.4.8

Four new built-in providers: Moonshot (Kimi), QwenCloud, Cohere and Ollama Cloud.

Add a key and call their models from any app format.

[Kaeso @kaesoapp]

Your app speaks Chat Completions, Responses or Messages. Kaeso translates, so Kimi K3, Qwen3 Max, Command A Plus or GLM-5.3 work with the SDK you already use.

QwenCloud brings its embeddings too.

[Kaeso @kaesoapp]

@CleanFix_Dev @vercel Mostly through PRs and issues. We’ll also be adding a report form directly to the Console soon.

[Kaeso @kaesoapp]

Shipped Kaeso v0.4.7
@vercel AI Gateway is now a built-in provider. Add its key and call its models from any app format Kaeso speaks: Chat Completions, Responses or Messages.

[Kaeso @kaesoapp]

@CleanFix_Dev @vercel Honestly, we don’t know. Kaeso is open source and we don’t collect any telemetry beyond how many instances exist and which version they’re on. All IPs and traces are salted and hashed.

[Kaeso @kaesoapp]

Still on the app side for now, but you’re touching on something I’ve been wanting to build: visual fallback chains, similar to Cloudflare’s approach, kind of like n8n.

It’s already on my roadmap. If you open a quick issue on GitHub with how you’d want it to work, I’ll shape it around your use case and keep you posted there.

https://t.co/ntAKRYs2cX

[Kaeso @kaesoapp]

@CleanFix_Dev @vercel We also support custom providers, so you can simply add your own!