What 'sillytavern summarize' Means in 2026

SillyTavern's summarize feature automatically creates a short summary of your ongoing conversation history, which replaces older messages in the context window. This prevents the total token count from exceeding your API model's limit (typically 4K–32K tokens depending on the provider). Without summarization, long roleplay sessions would either break mid-scene or force you to delete history manually. The summary is generated by the same AI model you're chatting with, using a dedicated prompt you can customize. By default, it condenses the last several messages into a paragraph or two, capturing key events, emotional states, and character actions. This allows you to continue conversations for hundreds of messages without losing context. The feature is enabled per-chat (or globally) and triggers automatically when the context approaches the token limit. You can also manually trigger a summarize at any time via the interface's "Summarize" button. The summarize output is stored as a separate entry in the chat history, so you can review or edit it if needed.

To enable summarize in SillyTavern, open the Extensions menu (puzzle piece icon) and toggle the 'Summarize' extension. Then navigate to the Summarize settings panel. Here you can set the 'Token budget' — how many tokens the summary should aim for (default 512 tokens). The 'Target length' slider controls how many messages back the system considers for summarization (default 10 messages). You can also adjust the 'Frequency' — how often automatic summarization runs (every N messages). For most users, leaving the default trigger at 'Context size threshold' works best: summarization fires when your context reaches ~90% of the model's limit. Advanced users can edit the summary prompt template under 'Prompt settings'. The default template instructs the AI to "Summarize the conversation so far, focusing on important events, character development, and current goals." You can tweak this to prioritize romance, mystery, or gameplay elements. Finally, choose whether to apply summarization to the current chat only or to all new chats globally. Changes take effect immediately with no restart needed.

The default SillyTavern summarize prompt works for general chat, but roleplay benefits from a tailored template. A popular community template is: "Write a concise summary of the roleplay so far. Include the setting, current location, key characters, their relationships, ongoing conflicts, and immediate goals. Use third-person past tense. Keep it under 200 words." For erotic roleplay (ERP), many users add: "Maintain a neutral tone. Do not include explicit details — only emotional intimacy, relationship progression, and consent boundaries." This prevents the summary from containing NSFW content that might trigger API filters. For mystery or investigation scenarios, try: "Summarize clues discovered, suspects encountered, and unresolved questions. Keep a timeline of events." You can also use the summary prompt to inject character traits: "Summarize from [Character Name]'s perspective, noting their feelings and observations." Save multiple templates as separate text files and swap them in the prompt field. Test each template with a 50-message chat to see which produces the most useful summaries. Avoid templates that generate overly long or flowery prose — concise summaries save more tokens.

Summarize can fail if the AI model rejects the summary prompt (e.g., due to content filters) or if the API returns an error. Common failure signs: the summary appears as a blank message, an error message in red, or the conversation stops summarization without notice. If using a free or low-cost API (like certain OpenAI tiers or proxies), token limits on the request itself may cause truncation. The fix: reduce the 'Token budget' in settings to 256 or even 128 tokens. Another issue: the model might generate a summary that exceeds the budget, which gets cut off mid-sentence. SillyTavern handles this by trimming to the nearest sentence boundary, but sometimes it fails. You can manually edit a truncated summary by clicking the edit icon on the summary entry. For OpenAI's ChatGPT models (GPT-3.5-turbo-16k and GPT-4-turbo), summarization works reliably. Claude models (Anthropic) sometimes refuse to summarize explicit roleplay; switch to a less restrictive model or use a custom prompt that omits details. If summarization stops working entirely, disable and re-enable the Summarize extension, then restart the chat.

SillyTavern offers two modes: automatic summarization (triggered by context threshold or message count) and manual summarization (via a button click). Automatic mode is hands-off but can produce summaries at awkward moments — right before a dramatic reveal, for instance, causing loss of tension. Manual mode gives you control: you decide when to summarize, allowing you to preserve key scenes. However, manual summarization requires you to remember to do it before hitting the token limit. A hybrid approach works best: set automatic summarization to trigger at 95% of the token limit (as a safety net), but manually summarize every 50–100 messages at natural breakpoints. Many experienced users disable automatic summarization entirely and rely on manual triggers, combined with periodic deletion of very old messages when needed. The choice also depends on your API model's context window: with 32K tokens (e.g., GPT-4-turbo or Claude 2.1), you can go hundreds of messages before needing any summarization. With 4K models (e.g., GPT-3.5-turbo), automatic summarization is almost mandatory after 20–30 messages.

If configuring API keys, managing token budgets, and tweaking prompt templates sounds exhausting, AIAngels offers a simpler alternative. AIAngels is an AI companion platform with permanent memory that never degrades — no summarization needed. Every conversation is stored in full, and the AI references back to any detail from day one, even after months of daily chat. There's no token limit, no context window to manage, and no summarization settings to tweak. For $12.99/month (or $2.99/month on annual plan), you get unlimited text chat, image generation, voice messages, and 70+ companions plus a custom companion builder. The free tier also offers unlimited text chat with no message caps — just basic memory (last 30 days). While SillyTavern gives power users granular control, AIAngels is built for people who want to talk to an AI without becoming a system administrator. If you're spending more time configuring SillyTavern than actually chatting, AIAngels might be a better fit.