Transparency
Every reply on Reverie comes from a model we reach through its vendor's official API or through OpenRouter. No reverse-engineered endpoints, shared accounts or resold keys.
Read from the routing code on every deploy · Chat models: 10
Unofficial relay Not on the path. Reverse-engineered endpoints, shared accounts and resold keys cost a fraction as much; we don't use them.
Nothing is added on the way. What reaches the model is your character, your chat and our own instructions; no relay wraps it in a system prompt of its own.
Who generated a reply
Credit center › Usage
Chat replies, story turns and novel chapters each show the service and the host that ran them.
Generated by
DeepSeek, via OpenRouter
A request goes to the first route; the next is tried only when it fails.
OpenRouter · then Vercel AI Gateway
An aggregator: each request runs on one of the hosts it lists for the model.
| Model | Providers it can run on |
|---|---|
| DeepSeek V3.2DeepSeek |
|
| DeepSeek V4.1 FlashDeepSeek |
|
| DeepSeek V4 ProDeepSeek |
|
| DeepSeek R1DeepSeek |
|
| GLM 5.3 FlashZ.AI |
|
| GLM 4.7Z.AI |
|
| GLM 5Z.AI |
|
| Gemini 3 Flash PreviewGoogle |
|
| Llama 3.1 8BMeta |
|
OpenRouter
An aggregator: each request runs on one of the hosts it lists for the model.
| Model | Providers it can run on |
|---|---|
| MiMo V2.6 FlashXiaomi |
|
OpenRouter lists the providers that serve each model. We remove the ones we exclude, and OpenRouter picks one of the rest for each request; your usage history shows which one it was.
OpenRouter can run one model on many hosts, and not all of them serve it well.
On top of these, a probe tests every provider every 30 minutes and suspends one that fails for a while.