Roleplay Engine

The Roleplay Engine is the part of Abolitus that turns your workspace into a coherent prompt.

That matters because the model does not see your app. It only sees the final context that Abolitus builds for the current turn. If that context is ordered badly, the model forgets who you are, ignores your lore, or wastes most of the window on low-value text.

What the Roleplay Engine Does

On every send, Abolitus rebuilds the prompt from the current workspace state.

That can include:

• Your selected persona.
• Character definition and scenario.
• Lorebooks from several scopes.
• Recent chat history.
• Author's Note.
• Prompt wrappers.
• Rolling summaries.
• Retrieved lore that matches the current scene.

The important point is that Abolitus does not treat every block equally. Some blocks are much more important than others and are protected accordingly.

The Practical Priority Order

From a user perspective, the engine behaves roughly like this:

1. Preserve scene identity first.
2. Preserve stable world rules second.
3. Preserve recent conversational momentum third.
4. Drop lower-value context before dropping high-value anchors.

That is why, in long sessions, example dialogue or older chat turns are more likely to disappear from active context than the core identity of the scene.

What Usually Stays Stable the Longest

These blocks are typically the most valuable:

• Persona.
• Character definition.
• Important lorebook entries.
• Active wrappers.
• Author's Note when enabled.

These blocks are what keep the model feeling like the same character in the same world even after the chat becomes long.

What Gets Squeezed First When Context Gets Tight

When the window is under pressure, Abolitus starts cutting from the parts that are easiest to sacrifice.

In practice, that usually means:

• Less essential older chat turns.
• Low-priority examples.
• Any block you explicitly exclude through prompt-pruning settings.

This is why long-term facts should live in personas or lorebooks rather than in one old message you hope the model will keep remembering forever.

Active Chat History vs Saved History

This is a very common source of confusion.

Saved history

Your old messages can still exist in the local database.

Active history

Only the portion that still fits the model's current context window is sent with the current turn.

If older messages are trimmed from active context, they are not necessarily deleted from your device. They are simply no longer part of what the model can read right now.

Why Author's Note Feels Strong

Author's Note is designed to influence the near-term output. It is injected close to the current turn rather than buried in distant history.

Use it for:

• Short-term tone corrections.
• Immediate pacing control.
• Temporary scene steering.
• Behavior reminders that matter right now but should not become permanent character law.

Do not use it for:

• Permanent world rules.
• Core character identity.
• Large encyclopedic lore blocks.

Those belong in the character card or lorebook instead.

Prompt Pruning Matters More Than People Expect

Prompt pruning decides which major blocks are allowed into the final prompt at all.

This is useful when you want to:

• Save tokens.
• Reduce clutter.
• Run smaller local models more cleanly.
• Test which memory source is actually helping.

For example, if a small local model becomes unstable under a very dense scene, trimming optional blocks can improve output more than switching samplers.

Rolling Summaries

When enabled, rolling summaries help preserve continuity from older material that no longer fits comfortably in the live context window.

Use summary modes when:

• Sessions are long.
• Plot continuity matters.
• You want fewer hard memory drops late in a story.

Keep them off when:

• You want the cleanest possible raw chat context.
• You are testing exact prompt behavior.

Best Practices

Put durable facts in durable places

Use personas for who you are, character definitions for who the other side is, and lorebooks for facts the scene should reliably recall.

Use chat for momentum, not archival memory

Chat history is important, but it is also the part most likely to be squeezed as a session grows.

Watch the token budget before a scene becomes huge

Do not wait until the story is already massive. Learn how your preferred route behaves under pressure early.

Use Author's Note for active steering only

It is for tactical influence, not for replacing the rest of your setup.