sillytavern tracker: The No-Setup Alternative

A SillyTavern tracker is any system—built-in, third-party, or custom—that records how many tokens your conversations consume, how much they cost (if using a paid API like OpenAI or Claude), and how often you interact with each character. SillyTavern itself does not include a native cost tracker; you must rely on external tools or manual logging. Without a tracker, users often blow through API budgets, especially when using long context windows or large models like GPT-4. A tracker also helps you identify which characters or conversation styles drive up token usage. For example, a single roleplay session with a 4K-token context can cost $0.06 on GPT-4, but with a tracker you can cap daily spend. Some users build simple Python scripts that parse SillyTavern's JSON logs to extract token counts per session. Others use browser extensions like Token Counter for OpenAI. The key is to track before you hit a surprise bill.

SillyTavern offers a few built-in indicators that can serve as a basic tracker. In the settings panel, you can view the current context token count—this shows how many tokens are in the active conversation window. However, this number resets when you clear context or start a new chat. There is no cumulative token log or cost calculator. The 'API Statistics' tab (available in some forks) displays total tokens sent and received per session, but it does not persist across restarts. For a persistent tracker, you must enable logging in the config.yaml file by setting `enableLogging: true`. This writes raw API request/response data to a local file, which you can then parse with a script. The built-in features are minimal; they give you a snapshot, not a history. If you want to track costs over days or weeks, you will need to supplement with an external tool.

Several third-party tools integrate with SillyTavern to provide detailed token and cost tracking. One popular option is the 'SillyTavern Token Tracker' extension available on GitHub. It adds a live token counter to the UI and logs usage to a local SQLite database. Another is 'ST-Cost-Calculator', a Python script that reads SillyTavern's log files and outputs daily/weekly cost summaries for OpenAI, Anthropic, or local models. For users running KoboldCPP or TextGen WebUI locally, tools like 'KoboldAI-Token-Counter' can be adapted. There are also browser-based solutions: the 'Token Counter' Chrome extension works with any webpage, including SillyTavern's interface, to estimate tokens in real time. However, these tools require manual setup and may break after SillyTavern updates. The most reliable approach is to use a dedicated API dashboard from your provider—OpenAI's Usage page, for example—but that lacks per-character granularity.

Building a custom SillyTavern tracker is straightforward if you have basic Python skills. First, enable logging in SillyTavern by setting `enableLogging: true` in `config.yaml`. This creates a `logs/` directory with JSON files containing each API request and response. Write a Python script that reads these files, extracts the `prompt_tokens` and `completion_tokens` fields, and sums them per session or per character. You can also multiply by your API's per-token cost (e.g., $0.01 per 1K tokens for GPT-3.5). Store results in a CSV or SQLite database for historical analysis. A sample script might use `os.listdir()` to iterate log files, `json.load()` to parse each, and `pandas` to aggregate. Add a simple Tkinter GUI or a Flask web interface to view stats. This approach gives you full control: you can filter by character name, date range, or model. The downside is maintenance—SillyTavern's log format can change between versions.

Tracking SillyTavern usage comes with several pitfalls. First, token counts from different providers are not directly comparable—OpenAI counts tokens differently than Anthropic or local models. Second, SillyTavern's context compression (e.g., 'Summarize' or 'Vector Storage') can obscure true token consumption because the summarization itself costs tokens. Third, if you use multiple API keys or rotate models mid-conversation, your tracker must handle that complexity. Fourth, local models (like Llama 2) have no monetary cost, but tracking tokens is still useful for performance tuning. Fifth, log files can grow large quickly—a heavy roleplay session might generate 10 MB of logs per day. Finally, privacy: logs contain your full conversation text. If you share your tracker or store logs in the cloud, you risk exposing sensitive data. Always encrypt logs or store them locally.

If SillyTavern's tracking complexity feels overwhelming, platforms like AIAngels offer a simpler alternative. AIAngels provides a built-in usage dashboard that shows your message count, image generations, and voice message usage without any setup. There are no API keys to manage, no token math, and no log parsing. For $2.99/month on the annual plan, you get unlimited text chat with 70+ companions, permanent memory, and image generation—all with transparent pricing. While SillyTavern gives you maximum control, AIAngels gives you maximum convenience. If your priority is tracking costs and usage without writing scripts, AIAngels handles it automatically. You can focus on the conversation, not the configuration.