Choose Your RoleCall Setup

RoleCall deliberately does not turn every system on for every scene. Each extra agent, memory writer, image pass, and post-production pass can add latency, token use, or another decision point. “Everything enabled” is not the best setup—it is simply the busiest one.

Pick the experience you actually want. You can save the result as a Scene Loadout and stop rebuilding it one switch at a time.

---

The Thirty-Second Answer

If you want…	Start with…	Leave off until needed
Fast, uncomplicated roleplay	Character + Preset + one narrator model	Story Director, Compendium writes, Trackers, Post-Production
Stronger prose control	Character + Preset + narrator model + selected Guides or one Post-Production pass	Campaign management and automatic memory writers
A long-running campaign	Character + Preset + Summary + Compendium + optional Story Director	Image automation and unrelated Trackers
Quality first; cost and latency are secondary	A premium narrator + premium specialist models + saved Loadout	Only systems unrelated to your story
Maximum privacy and control	BYOK provider + your preferred encryption mode + a saved Loadout	Managed-provider fallback, unless you want it

When in doubt, use Simple Scene below. Add a system only when you can name the problem it solves.

---

Setup 1: Simple Scene

Best for everyday chat, testing a new character, short stories, and anyone who does not want a cockpit full of switches.

Turn on

• A character and, optionally, a persona
• One preset
• One narrator model
• Streaming, if the selected provider supports it
• A sensible context budget that fits the model

Leave off

• Story Director
• Compendium automatic writes
• Trackers
• Post-Production agents
• Automatic image generation

Why

The narrator receives the cleanest prompt and begins answering immediately. Nothing is wrong when the other systems say Off—they are optional production departments, not missing setup steps.

Use the creator’s recommended preset and model when one exists. Avoid changing ten samplers at once; presets are designed as a coherent set.

---

Setup 2: Prose Workshop

Best when the character is behaving correctly but the writing style needs stronger control.

Start from Simple Scene, then add one of these

1. Guides for instructions that should apply to one generation or one kind of correction.
2. Author’s Note for a persistent rule that should ride with several turns.
3. Anti-Slop for recognizable banned phrases and repeated prose habits.
4. Scene Editor for a broader rewrite pass.

Do not enable every prose tool automatically. They can fight each other: a preset may ask for lush description while a post-production pass aggressively compresses it.

Recommended order

1. Fix the preset or prompt first.
2. Add a Guide for situational steering.
3. Add one Post-Production pass only if generated text still needs cleanup.
4. Inspect the final prompt and post-production receipt before adding another pass.

Cost and speed

Guides and Author’s Notes alter the narrator request and usually do not require a second model call. LLM Post-Production passes do require additional inference and therefore add time and token usage.

---

Setup 3: Continuity Campaign

Best for long-running roleplay where facts, relationships, inventory, and unresolved threads need to survive beyond the immediate context window.

Start from Simple Scene, then add

• Summary to compress older conversation history
• Compendium reading so established world facts return to the narrator
• Compendium writing when you want new durable facts collected automatically
• Story Director only when you want active pacing, planning, or encounter direction
• Only the Trackers that represent real game or story state you intend to maintain

Do not confuse these systems

• Summary compresses the conversation.
• Lorebooks inject authored world information when their triggers match.
• Compendium maintains structured continuity learned during play.
• Story Director plans and steers future beats.
• Trackers maintain a particular state domain.

They can cooperate, but they are not interchangeable. Enabling all of them does not create “more memory”; it creates several writers with different jobs.

Context guidance

Use the selected model’s real context capacity as the ceiling. A larger slider value cannot make a 32K model accept 128K. For many campaigns, a moderate recent-message window plus Summary and selective Compendium context is cheaper and clearer than filling the entire model window every turn.

---

Setup 4: No-Compromise Studio

Best when you care more about a dependable, high-quality result than minimizing model calls, tokens, or wait time.

Recommended departments

• A premium long-context model as the narrator
• A strong planning model for Story Director when you want active pacing
• Summary and Compendium reading for long-running continuity
• Compendium writing when you want durable facts collected automatically
• One premium Post-Production pass with a single, explicit job
• Image automation only for scenes where visual assets are part of the experience

Choose the strongest currently available model for each job rather than copying an old model name from a screenshot. Model catalogues and provider availability change; the task assigned to the model is the durable part of this setup.

The safe order

1. Make the narrator work correctly by itself.
2. Add Summary and Compendium continuity.
3. Add Story Director and confirm its visible intervention frequency.
4. Add one Post-Production pass and inspect its receipts.
5. Save the verified configuration as a Loadout.

This can involve several inference calls for one visible turn. Expect higher token use and more latency. “Premium everywhere” still does not mean “every switch on”: duplicate memory writers, conflicting prose passes, and unused Trackers add noise rather than quality.

---

Setup 5: Private BYOK

Best when you want to choose the provider, endpoint, model availability, and billing relationship yourself.

1. Follow Setting Up BYOK Providers.
2. Test the connection before attaching it to an important scene.
3. Select the exact saved connection—not merely a provider family with the same name.
4. Confirm the model supports the capabilities you plan to use, such as tools, images, or streaming.
5. Save the completed scene configuration as a Loadout.

Custom OpenAI-compatible endpoints vary. RoleCall can send a standard request, but a proxy may omit streaming, tool calling, image inputs, or provider-specific reasoning controls. Treat an unknown capability as unknown rather than assuming support.

---

Save It Once

After a scene behaves the way you want:

1. Open the in-scene Save Loadout control.
2. Review which panels will be captured.
3. Exclude live story state such as messages, HP, inventory, and tracker values.
4. Give the Loadout a purpose-based name such as Fast Character Test, Prose Workshop, or Long Campaign.
5. In Settings → Loadouts, optionally make it the account default.

A Loadout is RoleCall configuration. It does not copy the current story or secretly apply preset-local configuration maps.

---

Ask Orison

When Orison is available, ask:

Help me choose between Simple Scene, Prose Workshop, Continuity Campaign, No-Compromise Studio, and Private BYOK. Ask what I care about before recommending settings, and do not change anything until I confirm.

Orison can explain the controls in the context of the page or scene you are viewing. You remain in control of any setting change.

---

Troubleshooting the Setup

“Why is everything off?”

That is a valid baseline, not an error. A character, preset, and narrator model are enough for a scene.

“My responses became much slower”

Count the model calls. Story Director, automatic Compendium work, Trackers, and each LLM Post-Production pass may run separately from the narrator.

“The model ignores my style”

Inspect the effective preset and prompt before adding more systems. Conflicting Guides, Author’s Notes, preset prompts, and rewrite passes can cancel one another.

“My long chat forgot something”

Check the actual model context limit, Summary state, active Lorebooks, and Compendium read mode. Raising a UI slider beyond the model’s supported window cannot restore omitted context.

“A Loadout changed more than I expected”

Review its captured panels before applying it. Use the application review to keep, merge, or skip each category. Preset-local prompt configurations and Scene Loadouts are separate systems.