Persona and voice for AI agents. Designed. Versioned. Deployed.
Most AI agents are prompt-shaped voids. Sigil gives them a behavioral specification — character, voice, decision patterns — that persists across every deployment and every model update. Accessible to humans and autonomous agents alike.
from l8nt import Sigil
# Load your versioned agent persona
aria = Sigil.load("acme/aria", version="2.1")
# Use the generated system prompt with any model
print(aria.system_prompt)
# → "You are Aria, Acme Corp's customer success agent.
# You are direct but warm. You never hedge. You escalate
# with confidence, not apology..."Persona Architecture
A structured behavioral specification — not just tone guidelines. How your agent thinks, what it avoids, how it escalates, what it sounds like under pressure. The document that makes your agent recognizably yours.
Voice Calibration
From abstract brand values to concrete language patterns. Sentence structure, vocabulary constraints, hedging rules, escalation triggers. The translation layer between your brand guidelines and behavior your model can actually act on.
Versioned Registry
Every persona lives in the Sigil registry with a full version history. Load it by name. Roll back to any prior version. Know exactly what changed between your agent's behavior last month and today.
Agent-Native Access
The registry is built to be a runtime dependency, not just a handoff artifact. Autonomous agents will fetch their own personas at inference time — no human subscription required. Built for the agent economy.
How it works
Discovery
Stakeholder interviews, brand audit, existing agent review. We map your use cases, your edge cases, and the gaps between your brand guidelines and what your agents actually do.
Persona Specification
A behavioral document that defines who your agent is: how it communicates, what it avoids, how it handles ambiguity, where it draws lines. The source of truth for everything downstream.
Prompt Engineering
Translation from spec to production system prompt. Tested against your real scenarios. Iterated until the output matches the character on paper.
Registry & Handoff
Your persona enters the Sigil registry, versioned and documented. Pull it by name. Update it later. Roll back if something breaks. Full handoff — no ongoing dependency.
Runtime Access
Coming soonThe registry is not just a build-time artifact. Autonomous agents will fetch their own personas at inference time — no human subscription, no API key provisioning. The persona travels with the agent.
Start a Sigil engagement
Ready to give your agents a persistent identity? Get in touch and we'll scope your project.
Get in touchFAQ
What is Sigil?
Sigil is L8NTLABS' agent persona design methodology and hosted registry. We build the behavioral specification, voice guidelines, and production system prompt for your AI agent — then store it in a version-controlled registry you can pull from anywhere, with any model.
How is this different from writing a system prompt?
A system prompt is the output, not the process. Sigil starts with stakeholder interviews, brand audits, and scenario mapping across your real use cases. The resulting specification drives the prompt — not the other way around. It's the difference between a character and a costume.
What do I actually get at the end?
A behavioral specification document, a voice calibration guide with concrete linguistic patterns, a production-ready system prompt tested against your actual scenarios, and a versioned entry in the Sigil registry. Full handoff. No black box.
Can an autonomous agent access Sigil without a human subscription?
That's exactly where we're headed. The Sigil registry is being built to support agent-native access — autonomous agents fetching their own personas at runtime, paying per-use without a human in the loop. Early access clients will get priority when this launches.
What does it cost?
Persona design engagements start at $2,500. Full pricing and service tiers are on the consulting page.