IntenseRP Next v2

It's a local OpenAI-compatible API + desktop app that drives various web LLM chat UIs (via Playwright), so you can use those providers from SillyTavern and other clients without wiring up the official provider APIs. Slightly cursed yet surprisingly effective.

What is this? · Quick start · Client setup · Providers · Documentation · Releases · Issues · Discord Server

🎬 Preview Video

INTENSERPNEXT2PREVIEW.mp4

Important

This project is being archived and is no longer actively maintained. The code and docs remain available for historical/reference use. Provider websites can change without notice, so existing integrations may break. Use this only with services, accounts, and data you are authorized to access, and follow the applicable provider terms.

Welcome 👋

If you're here because you need a local bridge between SillyTavern and a browser-based provider chat, this repository may still be useful as reference. IntenseRP Next v2 drives supported provider web apps (like DeepSeek, Z.AI, Kimi, QwenLM, Perplexity, HuggingChat, and Google AI Studio) in a real browser, and re-exposes the browser session as an OpenAI-compatible endpoint.

It can mirror provider web UI controls such as reasoning toggles, search, file uploads, and model pickers where those controls are available to your account. The tradeoff is that it depends on third-party web pages, so it is inherently fragile and should be treated as a local personal-use tool.

Start here! 🎁

1. Download a release (see Releases) and run it (or run from source)
2. Click Start and log in when the browser opens
3. Point your SillyTavern client at http://127.0.0.1:7777/v1 (default) and pick the matching provider model IDs (deepseek-*, glm-*, moonshot-*, qwen-*, etc.)

And it's done! It should Just Work™️.

What is this?

IntenseRP Next v2 (sometimes shortened to "IRP Next v2") is a local bridge between:

• an OpenAI-style client (like SillyTavern), and
• a supported provider web app (DeepSeek, GLM Chat, Kimi / Moonshot, QwenLM, Perplexity, HuggingChat, or Google AI Studio)

Under the hood it:

1. Starts a local FastAPI server (OpenAI-compatible routes under /v1)
2. Launches a real Chromium session (Patchright/Playwright)
3. Logs in (manual or auto-login)
4. Captures the provider's streaming responses in the browser session
5. Re-emits them as OpenAI-style SSE deltas for your client

In normal human terms, it is a local compatibility layer between a desktop browser session and OpenAI-style clients.

Should you use it? 🎯

If you read this far, you probably have a use case in mind! But here's the objective truth:

It would work well for you if you:

• want a local bridge for provider web models
• prefer a clicky desktop app over a pile of scripts
• are OK with the occasional wait or hiccup (web apps change)

Not the best fit if you:

• need high throughput / parallel requests without using the experimental parallel modes
• want to run headless on a server
• need an actively maintained project with support guarantees
• want something that never breaks (that's perhaps the biggest caveat)

Note

1. Provider web apps change. When they do, a driver can break until it's updated.
2. By default, IntenseRP processes one request at a time (requests are queued). This is on purpose for a single live browser session.
3. This project is not affiliated with DeepSeek, ZhipuAI, SillyTavern, or any provider.

Why v2?

v2 is a full rewrite based on lessons learned from the original IntenseRP API (by Omega-Slender) and my own IntenseRP Next v1. The focus is less on a pile of features and more on making it sane to maintain and hard to break.

It's a more modular codebase with a Playwright-first browser-control approach, a better UI (PySide6), and a cleaner settings model, plus built-in update and migration flows.

If you want to compare, have a look:

Area	IntenseRP API / Next v1	IntenseRP Next v2
Backend	Python (Flask)	Python (FastAPI)
UI	customtkinter	PySide6 (Qt)
Automation	Selenium-based	Playwright (Patchright)
Response capture	HTML parsing	Structured browser response handling

Quick start

Tip

First launch can take a bit - v2 will verify/download its browser components.

Windows (recommended)

1. Download the latest intenserp-next-v2-win32-x64.zip from Releases
2. Extract it anywhere
3. Open the intense-rp-next folder and run intenserp-next-v2.exe
4. Click Start and wait for the browser to open

Linux

1. Download the latest intenserp-next-v2-linux-x64.tar.gz from Releases
2. Extract and run:

tar -xzf intenserp-next-v2-linux-x64.tar.gz
cd intense-rp-next
chmod +x intenserp-next-v2
./intenserp-next-v2

If it complains about missing libraries, you may need Qt6 deps installed on your system. The best way is to install the qt6-base package via your package manager, but if it doesn't stop you can just install the missing libs manually.

From source (for devs)

Requirements: Python 3.12+ (3.13 recommended)

git clone https://github.com/LyubomirT/intense-rp-next.git
cd intense-rp-next

python -m venv venv

source venv/bin/activate  # Linux/Mac
# or: venv\\Scripts\\activate  # Windows

pip install -r requirements.txt
python main.py

Connect SillyTavern (or any OpenAI-compatible client)

Once the app says Running (Port 7777):

Setting	Value
Endpoint	http://127.0.0.1:7777/v1
API	OpenAI-compatible chat or text completions
API key	Leave blank (unless you enabled API keys)
Model	Provider model IDs like deepseek-*, glm-*, moonshot-*, qwen-*, perplexity-*, huggingchat-*, or aistudio-*

Available model IDs (depends on provider):

• DeepSeek:
  • deepseek-auto (uses your IntenseRP settings)
  • deepseek-chat (forces DeepThink off)
  • deepseek-reasoner (forces DeepThink on, Send DeepThink follows your setting)
  • deepseek-expert-auto (uses your IntenseRP settings, but with Expert Mode enabled in the web UI)
  • deepseek-expert-chat (forces DeepThink off, with Expert Mode enabled in the web UI)
  • deepseek-expert-reasoner (forces DeepThink on, Send DeepThink follows your setting, with Expert Mode enabled in the web UI)
• GLM Chat:
  • glm-auto (uses your IntenseRP settings)
  • glm-chat (forces Deep Think off)
  • glm-reasoner (forces Deep Think on, Send Deep Think follows your setting)
• Kimi / Moonshot:
  • moonshot-auto (uses your IntenseRP settings)
  • moonshot-chat (forces Thinking off, Send Thinking off)
  • moonshot-reasoner (forces Thinking on, Send Thinking follows your setting)
• QwenLM:
  • qwen-auto (uses your IntenseRP settings)
  • qwen-chat (forces Thinking off, Send Thinking off)
  • qwen-reasoner (forces Thinking on, Send Thinking follows your setting)
• Perplexity:
  • perplexity-auto (uses your IntenseRP settings)
  • perplexity-chat (forces Reasoning off)
  • perplexity-reasoner (forces Reasoning on)
• HuggingChat:
  • huggingchat-auto (uses your IntenseRP settings)
  • huggingchat-chat (uses HuggingChat's default/off behavior)
  • huggingchat-reasoner (enables HuggingChat Thinking where available)
• Google AI Studio:
  • aistudio-auto (uses your IntenseRP settings)
  • aistudio-chat (forces Thinking off, Send Thinking off)
  • aistudio-reasoner (forces Thinking on, Send Thinking follows your setting)

Note: these IDs are behavior presets (modes). For providers that support actual model selection in their web UI (like GLM, QwenLM, Perplexity, HuggingChat, and Google AI Studio), the mode will still apply its behavior changes, but the real web UI model will be whatever you have selected in settings within the app.

If you change the port in Settings, update the endpoint to match (example: http://127.0.0.1:YOUR_PORT/v1).

Tip

If you don't want to use different provider names for model selection, enable Universal Model Names in Settings. This replaces all provider names with intenserp, so the above would become intenserp-auto, intenserp-chat, and intenserp-reasoner. With the exception of DeepSeek's Expert Mode presets, which would become intenserp-expert-auto, intenserp-expert-chat, and intenserp-expert-reasoner.

Quick troubleshooting 🧯

• Browser takes forever on first run: it may be downloading/verifying Chromium. Let it cook, then try again.
• Client cannot connect: confirm the app says Running, and the endpoint matches your port (http://127.0.0.1:7777/v1 by default).
• 401 Unauthorized: you probably enabled API keys in Settings. Either disable them or add a key in your client.
• Login loops / stuck sign-in: try disabling Persistent Sessions, or clear the profile in Settings (it wipes saved cookies).
• Slow responses: requests are queued by default, and reasoning/thinking modes can add extra time. Experimental parallel modes can help, but they are much heavier.

Tip: enable the console and/or logfiles before reporting issues. Logs help a lot when diagnosing!

What you get ✨

There are a few highlights I think are worth calling out. Most have been in v1 as well, but v2 has them all better and in a cleaner way.

• 🖥️ A desktop UI that starts/stops everything for you (and doesn't require terminal work)
• 🔌 An OpenAI-compatible API under /v1 for SillyTavern and other OpenAI-compatibles
• 🧩 A formatting pipeline: templates, divider, injection, name detection
• 🧠 Provider behavior toggles: Thinking/Reasoning, Search, uploads, model pickers, and provider-specific knobs
• 👥 Saved Accounts, account rotation, pinning, disabling, and persistent browser sessions
• 🔐 Optional LAN mode and API keys
• 🪵 Built-in extensive logging: console window, log files, console dump
• 📱 Experimental Remote Control for managing the app from another device on your network
• ♻️ Built-in v1 migrator + built-in update flow (when running packaged builds)

Provider support

Current supported providers:

Provider	Status	Notes
DeepSeek	Stable	DeepThink, Search, Expert Mode, uploads, and chat reuse
GLM Chat (Z.ai)	Stable-ish	Deep Think, Search/Advanced Search/Tools, model selection, and CAPTCHA login
Kimi / Moonshot	Stable-ish	Thinking, Search, uploads, Google login, and Kimi-specific caveats
QwenLM	Stable	Thinking, Web Search, model selection, token counting, and uploads
Perplexity	Verification	Model selection, Thinking/Search controls, uploads, and email-code login
HuggingChat	Early integration	Model/provider selection, Thinking Effort, Exa search, uploads, and tight monthly credits
Google AI Studio	Usable	Gemini model selection, Thinking Level, Search/URL Context, and so on

More detail lives in docs/ (best viewed as the docs site - see below).

Documentation

There is a full docs site with screenshots and details if you want to dig a bit deeper:

Check out the docs site here.

Local preview (Zensical):

python -m pip install -r docs/requirements.txt
zensical serve

Project status

IntenseRP Next v2 is being archived.

Existing releases and docs remain available for people who already use the project or want to learn from the code. Community links may remain online, but no active maintenance or support response is promised.

Security and privacy notes

• IntenseRP is designed for local or LAN use. Do not expose it to the public internet unless you know what you're doing.
• If you enable Available on LAN, consider enabling API Keys too.
• Your config directory contains sensitive data (credentials, API keys, session cookies). Treat it like a password vault.

Contributing 🤝

This repository is being archived, so new issues and PRs may not be reviewed.

Existing issues, release notes, and docs remain available as reference. If you fork or continue the project privately, please review the provider terms, security assumptions, and maintenance risk before relying on it.

Contributors ❤️

LyubomirT	Omega-Slender	Deaquay	Targren	fushigipururin	Vova12344weq
Project Maintainer	Original Creator	Contributor to OG	Feedback & Proposals, Code	Code and Concept Contributor	Early Testing, Bug Reports, Suggestions

Full list: https://github.com/LyubomirT/intense-rp-next/graphs/contributors

License

IntenseRP Next v2 is licensed under the MIT License. See the LICENSE file for details.

Note

Original IntenseRP API by Omega-Slender is also MIT-licensed, but previously was a CC BY-NC-SA 4.0 project. This v2 rewrite is a new codebase and is not a derivative work, so the license has been switched to MIT for simplicity. I'm not affiliated with Omega-Slender, even if I'm the official successor to their project (starting from v1).

Credits

• FastAPI, Pydantic, Uvicorn
• PySide6 (Qt)
• Playwright + Patchright
• Feather Icons / Lucide Icons
• SillyTavern (client ecosystem)
• IntenseRP API (Omega-Slender) - original inspiration
• Me (LyubomirT) - for doing all the work :D
• RossAscends (for STMP)
• Developers of Zensical (docs generator)