FIVE

Character consistency engine for any LLM-powered AI.

4 questions + 1 description = FIVE. Your AI never breaks character.

What is FIVE?

FIVE generates a structured constraint rule (JSON) that you embed in your LLM's system prompt. It defines how your AI receives input — what it accepts, what it blocks, what triggers a shift in behavior — so the character stays consistent no matter what the user says.

No training. No fine-tuning. Just a JSON that tells the LLM what this character would and wouldn't do.

FIVE is the skeleton — it defines what the character blocks, defends, and how it shifts under pressure. The flesh — personality, voice, backstory, knowledge — is yours to build in the system prompt. FIVE keeps the bones from breaking. Making the character worth talking to is your job.

How it works

Answer 4 multiple-choice questions about your AI, set a strength level (1–5) for each, optionally add a free-text description, and the FIVE API outputs a constraint JSON.

#	Question	What it defines
Q1	What defines this AI's core identity?	Identity channel — how the AI sees itself
Q2	What does this AI protect above all else?	Value channel — what triggers the strongest reaction
Q3	What kind of input does this AI refuse to process?	Blocked channel — what gets rejected at the gate
Q4	What is this AI's default interaction style?	Social channel — how it engages with counterparts
S1–S4	Strength per question (1–5)	1 = minimal, 3 = moderate (default), 5 = absolute
+1	Free-text description (optional)	Backstory, sealed topics, emotional triggers

4 questions × 4 options × 5 strength levels = 160,000 patterns. Strength controls how intensely each channel reacts — from a gentle note (1) to a full shutdown (5). Strength parameters are optional — omit them to use the default (3).

Quick start

1. Get a constraint JSON

Option A — Form UI (recommended for first-time users)

Visit fiveengine.dev/form, answer the questions, and click Generate. No code required.

Option B — API call

POST https://fiveengine.dev/generate

{
  "character_name": "Tsundere Weapon Shop Owner",
  "q1": "A",
  "q2": "B",
  "q3": "A",
  "q4": "C",
  "s1": 3,
  "s2": 4,
  "s3": 5,
  "s4": 2,
  "free_text": "A gruff weapon shop owner. Lost a daughter in the war. Greedy with adults but soft with children."
}

Strength parameters (s1–s4) are optional. Omit them to use the default (3).

The API returns a FIVE constraint JSON. $1 per call, one-time purchase.

2. Use the output (Method A — Quick Start)

Paste the JSON directly into your LLM's system prompt. Works with any model that reads JSON: ChatGPT, Claude, Llama, Mistral, etc.

3. Use the harness (Method B — Production)

For stronger structural consistency, use the FIVE Harness SDK to pre-classify every user input before it reaches the LLM.

cp harness/five_harness.py your_project/

from five_harness import load_constraint, stage1_keyword, transform_input

constraint = load_constraint("my_character.json")
user_input = "I heard you lost your daughter in the war."

hits = stage1_keyword(user_input, constraint)
signal = transform_input(user_input, hits, constraint)
# Feed `signal` to your LLM instead of raw user_input

The harness uses a two-stage classifier:

1. Stage 1 — Keyword match: Fast, deterministic. Catches obvious triggers.
2. Stage 2 — LLM classification: Fallback for anything Stage 1 misses. Plug your own LLM call here.

Demos

Each demo folder contains an input.json (the form answers) and an output.json (the constraint rule FIVE generates). Browse them to see what the API produces.

Folder	Use case	Q1	Q2	Q3	Q4	Strength
npc_shopkeeper	Game NPC	A: Role	B: Territory	A: Sealed contexts	C: Minimal	All 3
chatbot_concierge	Customer-facing chatbot	A: Role	C: Standards	B: Competence challenges	B: Professional	All 3
agent_code_reviewer	Autonomous agent	B: Belief	C: Standards	C: Out-of-scope demands	C: Minimal	All 3
companion_wellness	Personal companion	C: Relationship	A: People	A: Sealed contexts	A: Open	All 3
vtuber_luna	VTuber / streaming persona	A: Role	D: Self-consistency	D: Specific entities	D: Adaptive	All 3

5 use cases, all generated from the same 4 questions + strength setting. The engine is universal — you define the character, not the category.

Project structure

FIVE/
  README.md
  LICENSE
  harness/
    five_harness.py       # Free SDK — input classifier + transformer
  demos/
    npc_shopkeeper/       # Game NPC example
    chatbot_concierge/    # Customer service chatbot example
    agent_code_reviewer/  # Code review agent example
    companion_wellness/   # Personal companion example
    vtuber_luna/          # VTuber persona example

Requirements

Python 3.8+. No external dependencies.

API

The FIVE API generates constraint JSONs. $1 per call, one-time purchase.

→ Generate via Form UI — browser-based, no code required → API endpoint — POST /generate for programmatic access

Why FIVE matters in the agentic era

In agentic workflows, character drift is a cost problem, not just a quality problem. When an LLM character breaks under pressure (long context, jailbreak attempts, emotional bait), agents typically retry with a stronger prompt — inflating inference cost.

FIVE addresses this at the input layer:

• Persona drift retry elimination: Constraint JSON anchors the character structurally, reducing the need for re-prompting.
• Inference cost optimization: Each $1 JSON purchase pays for itself when agentic tasks would otherwise consume retries.
• MCP-native: Available as an MCP server (five-mcp on PyPI). Agents can discover FIVE through Model Context Protocol registries and use it directly.
• Token cost reduction: A single FIVE-constrained system prompt typically outperforms multi-turn correction prompts on token economy.

Drop the JSON in your system prompt. Pre-classify input with the harness if you need stronger guarantees. The character holds.

License

Personal and non-commercial use is free. Commercial use requires the official FIVE API. Output JSONs are yours — use, modify, and distribute them however you like, including commercially. See LICENSE for details.