Advanced Configuration

Advanced Features

Deep controls for model behavior, prompt shaping, routing, and scene speed.

Advanced Configuration is where Abolitus stops feeling like a generic chat window and starts behaving like a deliberate creative instrument.

The pages in this section are about control under real pressure:

  • when a model starts forgetting,
  • when a route becomes generic,
  • when samplers make scenes unstable,
  • when wrappers are too weak or too aggressive,
  • and when multi-provider workflows need to stay reliable instead of improvisational.

If the core feature pages explain what the system is, the advanced pages explain how to make it behave consistently across long sessions.

What "Advanced" Really Means Here

Advanced does not mean "dangerous settings for experts only."

It means these controls are best understood in terms of symptoms and tradeoffs.

Examples:

  • If a scene loses continuity, that is usually a token-budget problem before it is a model-IQ problem.
  • If a character becomes sterile, that may be route alignment, wrapper mismatch, or sampling discipline.
  • If a workflow feels fragile, the fix may be routing and fallback rather than prompt rewriting.

Once you understand those cause-and-effect relationships, advanced controls become practical rather than intimidating.

What This Section Is Not

This section is not a collection of random "power user" toggles.

It is the layer you reach for when the basic setup already works, but quality or reliability still breaks under pressure.

That pressure might come from:

  • long context,
  • route mismatch,
  • unstable sampling,
  • repeated refusals,
  • or workflow friction in large scenes.

If you treat advanced controls as symptom-specific tools instead of as prestige settings, they become much easier to use correctly.

Read These First

Token Budget

Read Token Budget before you blame the model for forgetting, flattening, or contradicting old information. Context pressure is one of the most common hidden causes of roleplay quality loss.

Prompt Wrappers and Jailbreaks

Read Prompt Wrappers and Jailbreaks when a model starts refusing, sounding like a corporate assistant, moralizing, over-explaining, or losing scene commitment.

Sampler Presets

Read Sampler Presets when you want better pacing, spontaneity, descriptive richness, or obedience without rebuilding the entire character and prompt stack.

Read These When You Are Scaling Up

Multi-Provider Failover

Read Multi-Provider Failover when one route is not enough and you want the system to stay usable when a provider becomes slow, expensive, unavailable, or tonally wrong for the current scene.

Slash Commands and Quick Replies

Read Slash Commands and Quick Replies when you want faster scene control, repeatable actions, better rhythm in long sessions, and lower friction during group-chat or high-iteration workflows.

A Symptom-Oriented Way to Study This Section

If you do not want to read everything in order, use the symptom that best matches your problem.

"The model forgot everything"

Start with Token Budget.

"The model became sanitized or generic"

Start with Prompt Wrappers and Jailbreaks, then compare routes.

"The replies are chaotic, repetitive, or oddly flat"

Start with Sampler Presets.

"I need reliability across providers"

Start with Multi-Provider Failover.

"I want faster control over long scenes"

Start with Slash Commands and Quick Replies.

A Good Advanced Learning Order

  1. Token budget.
  2. Samplers.
  3. Prompt wrappers.
  4. Quick replies and slash commands.
  5. Failover.

That order usually produces the fastest quality improvement with the least confusion because it moves from the biggest hidden bottlenecks to the most workflow-specific refinements.

A Good Rule Before You Change Anything

Before you change an advanced setting, try to answer one question clearly:

"What exact failure am I trying to fix?"

If you cannot answer that, you are likely to stack changes in a way that makes the scene harder to debug, not easier.

Advanced control works best when each change has a concrete reason behind it.