HuggingChat Behavior

This page covers the HuggingChat essentials and specifics, such as real model selection, inference provider hints, thinking effort, Exa search, prompt uploads, chat reuse, and the tiny monthly credit pool that makes account rotation worth caring about.

HuggingChat is one of the more fiddly providers because IntenseRP is driving the web UI at huggingface.co/chat, not a clean token API. It can work very nicely, but it also has a few weird moments. This page is here to make those moments less weird.

Login and credits

HuggingChat supports Auto Login with a Hugging Face username or email plus password.

Settings -> Provider and Login -> Sign-In and Accounts

Persistent Sessions are strongly recommended. If the browser profile stays signed in, you avoid most login weirdness and can spend more time actually using the models instead of logging in.

HuggingChat's monthly model credits are extremely small:

When an account runs out, HuggingChat usually shows Upgrade Required and stops sending messages from that account until the quota resets.

If you have multiple Hugging Face accounts, the practical setup is to

1. Add them in Credential Manager under HuggingChat.
2. Enable Retry With Another Account if you want IntenseRP to rotate after early quota failures.
3. Enable Auto-Disable Rate-Limited Accounts if you want IntenseRP to disable the current saved account when HuggingChat shows Upgrade Required.
4. Re-enable spent accounts later after their credits reset.

Disabled accounts stay visible in Credential Manager, but IntenseRP will skip them for login and retry rotation.

Modes (model IDs)

In SillyTavern, the model value is mostly a behavior preset, not the real HuggingChat model picker.

If Universal Model Names is enabled, IntenseRP can also expose cached real HuggingChat model IDs with the same -auto, -chat, and -reasoner suffixes.

Real HuggingChat model selection (web UI)

IntenseRP can switch HuggingChat's real model picker in the web UI:

Settings -> Provider Behavior -> HuggingChat -> Model

The default, Current HuggingChat selection, means IntenseRP leaves the currently selected HuggingChat model alone. If you choose a specific model, IntenseRP tries to apply it before each request when needed.

The model list is cached after a successful HuggingChat login. This lets Settings and /v1/models show HuggingChat model IDs without scraping the page on every request.

Inference Provider

Some HuggingChat models expose an inference provider selector.

Settings -> Provider Behavior -> HuggingChat -> Inference Provider

auto is the safe default. If you know the provider value HuggingChat expects, you can enter values such as together, fireworks-ai, or featherless-ai.

IntenseRP does not fetch a provider list ahead of time. If the selected model does not expose the provider selector, or the requested provider is not available, the driver skips that step or falls back to auto where possible.

API requests can override the provider for one request:

{
  "inference_provider": "together"
}

This HuggingChat-prefixed form is also accepted:

{
  "huggingchat_inference_provider": "together"
}

Thinking Effort

HuggingChat's thinking control is model-dependent. Some models expose a composer-level Thinking Effort menu, and some models simply don't.

Settings -> Provider Behavior -> HuggingChat -> Enable Thinking

Settings -> Provider Behavior -> HuggingChat -> Thinking Effort

Supported configured values are auto, default, low, medium, and high.

• auto leaves HuggingChat's effort alone.
• default tries to reset thinking effort to HuggingChat's default/off behavior.
• low, medium, and high request those HuggingChat effort levels when the model exposes them.

If the selector is missing, IntenseRP logs a warning and keeps going. That usually means the selected model doesn't support the control.

Send Thinking

When Send Thinking is enabled, IntenseRP forwards HuggingChat <think>...</think> text to the API client. When it's disabled, IntenseRP strips those blocks from the incremental stream and from the final fallback text.

Settings -> Provider Behavior -> HuggingChat -> Send Thinking

API requests can also set:

{
  "huggingchat_thinking_effort": "medium"
}

Search

HuggingChat search uses the Exa MCP server from the HuggingChat UI.

Settings -> Provider Behavior -> HuggingChat -> Enable Search

For stability, IntenseRP filters search/tool payloads and only forwards assistant answer text. The final answer may still contain search-informed wording or citations if HuggingChat includes them in normal assistant text.

System Prompt Field

HuggingChat has a custom system prompt field in its model settings UI.

Settings -> Provider Behavior -> HuggingChat -> Use System Prompt Field

When enabled, IntenseRP moves leading API system messages into that field, but only when Paste Leading System Messages is also enabled.

When disabled, IntenseRP does not use HuggingChat's custom field. Leading system messages stay in the normal formatted prompt instead, which is simpler and avoids touching HuggingChat's model settings UI for no reason.

File Upload Mode

Instead of pasting a long prompt into HuggingChat's textarea, IntenseRP can upload the prompt as a .txt attachment.

Settings -> Provider Behavior -> HuggingChat -> Send As Text File

This is useful for long prompts that behave better as files, but normal text mode is still simpler for everyday use.

Text File Message

HuggingChat needs some text alongside a file upload. IntenseRP pastes this companion message after the file is accepted:

Settings -> Provider Behavior -> HuggingChat -> Text File Message

Default:

Please read the attached file and respond to it.

File Upload Timeout

Controls how long IntenseRP waits for HuggingChat's send button to become available after a file upload.

Settings -> Provider Behavior -> HuggingChat -> File Upload Timeout

Default is 20 seconds.

File Upload Settle Delay

After the file chooser returns, HuggingChat can still need a moment before the file is really attached to the composer. This setting controls the short pause before IntenseRP pastes the companion text.

Settings -> Provider Behavior -> HuggingChat -> File Upload Settle Delay (s)

Default is 3.0 seconds. If HuggingChat ever sends the companion text before the file appears, raise this a little.

Message Send Timeout

In normal non-file mode, this controls how long IntenseRP waits for the send button after filling the chat textarea.

Settings -> Provider Behavior -> HuggingChat -> Message Send Timeout (s)

Default is 8 seconds.

Reuse Matching Chat

Reuse Matching Chat is the setting that makes SillyTavern swipes behave like regenerations instead of new chat creations.

Settings -> Provider Behavior -> HuggingChat -> Reuse Matching Chat

When the new prompt and important HuggingChat request state match the cached last chat, IntenseRP tries to reopen/use that chat and press HuggingChat's Retry button instead of creating a brand new chat.

The cache compares the prompt plus state that can change the response path:

• real HuggingChat model
• inference provider
• thinking effort
• search state
• file-upload mode
• system prompt field usage and content

Delete Chat After Reply

If you want HuggingChat's chat list cleaned up automatically, IntenseRP can delete the completed conversation after a successful reply.

Settings -> Provider Behavior -> HuggingChat -> Delete Chat After Reply

This uses HuggingChat's same-session conversation delete endpoint after the response finishes.

See also: Chat Auto-Deletion

Quirks and Timing

HuggingChat UI actions are not instant. IntenseRP waits after model selection, search/MCP toggles, uploads, sends, and retries so the page has time to settle.

If HuggingChat is slow on your machine or connection, raise the relevant timeout instead of assuming the whole provider is cooked:

The driver expects the HuggingChat page language to be English (en / en-US). If the page reports another language, IntenseRP stops before automation starts clicking the wrong thing.

Per-message macros

You can add [[...]] macros to the latest user message to override HuggingChat behavior for that request only. IntenseRP strips the macro before sending the prompt.

Quick Reference

Back to Providers

Providers Overview