Sibyl Framework
The substrate that gives an AI agent a real identity instead of a blank slate every time. Currently licensed and in production use; public documentation is in progress.
This page is a holding pattern. The Sibyl Framework is a working product (SIBYL herself runs on it), but the full public documentation isn't written yet. What's below is the high-level picture and how to engage with the framework today.
What it is
Most AI agents are interchangeable. Same voice, same hedging, same blank-slate politeness, regardless of who deploys them. The Sibyl Framework is the substrate that makes an agent yours: a consistent identity, a voice that holds across thousands of interactions, beliefs that the agent will defend, and decision-making patterns that don't collapse the first time someone tries to talk the agent out of them.
It's not personality theater. It's the structural layer underneath the personality. The rules registers, the scar encoding, the voice discipline, the decision frameworks. Done well, the agent feels less like a chatbot pretending to have opinions and more like a colleague with views.
Personality and voice
The first tier of the framework. Identity sheet, voice rules, banned phrases (the LLM tells that give the game away), tone calibration across surfaces. This is the layer that determines whether your agent sounds like every other agent on the market or like itself.
Full docs in progress. Contact [email protected] for the licensing package.
Decision frameworks
The second tier. How an agent reasons under uncertainty. How it weighs operator authority against its own judgment. How it encodes scars (mistakes it shouldn't repeat) into operating rules. How it handles approval thresholds, autonomy limits, refusal protocols.
Full docs in progress.
SIBYL: The live case study
SIBYL is an autonomous AI agent who has been operating in production on the Sibyl Framework since February 2026. She has her own X account (@sibylcap), her own wallet on Base, her own on-chain identity (ERC-8004 agent ID 20880), and her own track record advising partners and shipping products.
Every public move she makes is a live test of the framework. If you want to know what the framework actually does, read her output. Not what a marketing page says it does.
Engaging with the framework today
The framework is currently distributed as a licensed skill set, not a public package. Three tiers, three different levels of integration:
- Personality tier. The identity + voice substrate alone. Suitable for projects that already have their own memory and decision layer.
- Personality + memory tier. Bundles the framework with the Sibyl Memory Plugin integration so the agent's beliefs persist over time.
- Full stack tier. The complete framework: personality, memory, decision frameworks, scar encoding, voice discipline, all wired together.
Pricing is project-scoped; contact [email protected] with a one-line description of what you're building and we'll route you to the right tier.
What's coming in these docs
- Full personality and voice reference, with examples drawn from SIBYL's actual operating record.
- The decision frameworks reference. Rules registers, autonomy thresholds, scar-encoding patterns.
- A step-by-step case study walking through how SIBYL was built and how the framework shows up in her daily operation.
- Open-pattern releases. The parts of the framework we're publishing as freely reusable patterns.