Use Arscelare 1 in SillyTavern
On this page
You can use Arscelare 1 for roleplay in SillyTavern without running a model on your computer. Keep using SillyTavern for your character cards and chats. Arscelare generates the replies.
Check for a 5 USDC trial spot. If a spot is available and you’re eligible, you can try a scene before paying. If you want to keep writing with Arscelare, you can top up with crypto.
What goes to Arscelare
SillyTavern sends the context it assembles for each request to Arscelare. That can include character instructions, lorebook entries and earlier messages, as well as what you just wrote.
We don't store your prompts or replies, and we don't read your conversations. Generation and input safety checks process text automatically. We keep account, payment and usage records, including token counts and costs. The Privacy page explains those records.
Arscelare is for adults aged 18 and over. Adult fictional content is supported; sexual content involving minors is prohibited. The model or safety checks can still refuse a request. Read the Terms before buying credit.
Connect through Chat Completion
Once you have credit, create an API key and open the SillyTavern setup guide. It has the exact URL and model ID, plus the steps for sending your first message.
Give it a scene you know
Choose a character whose voice you can recognize. Start a short chat and give the character something specific to respond to. You can use your existing card; you don't need to write a new one to evaluate a model.
Pay attention to more than whether the first reply sounds polished:
- Does the dialogue fit the character?
- Does the model leave your character's decisions to you?
- Does it keep track of who is present and what has happened?
- Can you steer the amount of description and the pace of the scene?
If you want a shorter reply, give an instruction about the writing itself. For example:
Reply as your character in two short paragraphs. Leave my character's actions and decisions to me.
This is an example instruction, not a guarantee of a particular length. Judge the result over several turns. If you change your preset, alter one part at a time so you can tell what affected the reply.
Reasoning shares the reply's token allowance
Arscelare uses medium reasoning by default. Reasoning tokens count toward both the output allowance and the bill, even when SillyTavern doesn't display them.
A low maximum-output setting can leave little room for the visible reply. If the model uses the allowance while reasoning, the response can contain no visible answer and still incur a charge.
SillyTavern's reasoning switch may not send a reasoning-effort value for a custom model named ars-1. A request that omits that value uses Arscelare's medium default. The API setting reasoning_effort: none disables reasoning; check the model documentation rather than assuming a client switch has changed it.
Your preset may also send sampling values that override server defaults. The model documentation lists the defaults Arscelare uses when a request leaves those settings out.
What a reply costs
Your request can contain much more than your latest message. Character instructions, a large preset, lorebook entries and the history SillyTavern sends all contribute to input usage.
| Usage | USDC per million tokens |
|---|---|
| Input | 1.50 |
| Cached input, when the server reports a cache hit | 0.375 |
| Output, including reasoning | 3.00 |
For an illustrative request with 10,000 uncached input tokens and 1,000 total output tokens:
| Part of the request | Cost |
|---|---|
| 10,000 input tokens | 0.015 USDC |
| 1,000 output tokens | 0.003 USDC |
| Total | 0.018 USDC |
That is arithmetic for those assumed token counts, not a measured average conversation. A request with 40,000 uncached input tokens and the same output would cost 0.063 USDC. Regenerating a reply sends another request and adds usage.
Caching can reduce the input charge when the model server reports cached tokens. Using the same character or preset does not guarantee a cache hit. Check your usage costs after the first few replies and use the current pricing page before adding more credit.
If a reply fails or cuts off
A connection error: recheck the base URL, API key, ars-1 model ID and credit balance. Arscelare does not support tool calls. If a preset includes tools, choose a prompt post-processing option without them. Don't bypass the API status check unless Test Message succeeds and only the status indicator is wrong.
A context-limit error: Arscelare allows 65,536 tokens for the fully rendered prompt and requested output allowance together. Maximum output is 8,192 tokens. If you omit an output limit, the server reserves that full allowance. Arscelare rejects an oversized request rather than trimming your messages or reducing the requested output. Check the history, preset and lorebook content your client sends.
A cut-off or empty reply: check the output allowance and reasoning setting. The token ceiling limits reasoning and visible text together. Lowering it does not ask the model to finish its prose more concisely.
If you need help, email support@arscelare.ai with your client version, selected API type and error message. Keep API keys and private conversation text out of the message.
Connect your character to Arscelare 1
Check trial availability. Once you’re connected, give Arscelare the next turn in a scene you know well.