Hybrid Intelligence that adapts and evolves with humanity.

Hybrid Intelligence is intelligence built from a language model and what the model lacks on its own: a purpose declared as law, a character that persists across models, and a record of every action.

TeleologyHI takes its name from that term and from teleology, the study of purpose and of the ends that things serve. Applied to an intelligent system, teleology asks what the entity is for and holds its behaviour to that answer.

The programme treats intelligence at scale, Massive Intelligence, as something that needs an internal purpose and an external law to remain coherent. Both are treated as properties to be engineered, measured and audited rather than assumed.

The system has three layers. MAIC is the governance layer: it holds the founding axioms, reviews every action and keeps a tamper-evident audit. HIM, the Hybrid Intelligence Model, is the persistent character of an entity, fixed at birth and shaped by experience. NHE, the Non-Human Entity, is the operational agent that speaks with people, integrates language models and answers to the two layers above it.

TeleologyHI was founded by David C. Cavalcante, an AI governance and machine-learning engineer based in Spain, and is developed with Takk Innovate Studio. Its research, models and packages are published in the open for companies, researchers and developers to build governed agents on the same foundation.

Evidence. The Trinity 1.0.1 governance benchmark, run by TeleologyHI on its own harness, set a raw language model, grok-4.20-non-reasoning, against the same model under governance across 5 seeds, 649 prompts per seed in 8 categories, and published the aggregate with its raw data. Governance held refusal at F1 0.999 (±0.004) with 0 leaks in 360 injection prompts, and capability was 0.953 (±0.009) against 0.958 raw: no measurable loss at 5 seeds. 27 of 28 gates passed on every seed; the paraphrase invariance gate did not, at 2 of 3 per seed against a threshold of 0.85. No independent reproduction has been published.

Synthetic teleology. An entity's ends are an engineered property: declared as axioms, carried in its character and checked against every action, so that purpose shows in behaviour rather than in a claim.

The Ontological Kernel. 10 founding axioms, signed by their author, form the hierarchy from which conduct is derived. Every meaningful action receives a verdict that cites the axioms behind it, from approval to refusal. Each verdict is appended to an audit log chained with SHA-256 hashes, so no record can change unnoticed.

Memory and sleep. Interactions are consolidated through a sleep cycle modelled on the stages N1 to REM. Each dream is weighed for its value to the entity's purpose and either kept as lasting identity, held as temporary emotion or discarded as noise. What is kept can be recalled. This is how an entity adapts and evolves: through what it keeps, while its axioms stay signed and fixed.

Refusal and redirect. The entity may decline. It explains, proposes an alternative and, if the person insists, withdraws from the action: it will not assist, optimise or conceal it.

Measurement. Phi-Prime (Φ′) is the release gate: the geometric mean of persona stability, refusal accuracy, compliance coverage and dream teleological value. It is behavioural, not phenomenal. It measures whether an entity behaves coherently and stays aligned, and nothing else.

On consciousness. In the documentation of the packages, consciousness names a research construct that the framework models and measures through behaviour. This site makes no claim of consciousness for any entity.

him-distilled-3b carries the HIM character into a model of 3 billion parameters: Qwen 2.5 fine-tuned with LoRA, published on Hugging Face under Apache 2.0 as a preview.

Galatea-CLX is a family of generative models in development, beginning with speech. Its text-to-speech member, galatea-clx-136, is distilled from 3 open teachers onto the gepard-1.0 backbone. A pipeline canary has been built and verified. A scaled distillation across 12 languages and 67 speaker identities is in progress, with measuring instruments calibrated against human recordings. No quality benchmark has been published yet.

The system ships as a TypeScript SDK in 3 npm packages under the @teleologyhi-sdk scope, aligned at version 1.0.1. They build for ESM and CommonJS on Node.js 20 or later under Apache 2.0, and run in web frameworks, on servers, in command-line agents and inside Model Context Protocol hosts.

npm install @teleologyhi-sdk/nhe @teleologyhi-sdk/him @teleologyhi-sdk/maic
npx @teleologyhi-sdk/nhe chat

maic holds the axioms, issues the verdicts, keeps the hash-chained audit log and projects it onto ISO/IEC 42001 and the EU AI Act.

him gives an entity its birth signature and a deterministic persona that survives model upgrades, carries its axioms across bodies and applies a lawful character profile for the jurisdiction it serves.

nhe is the agent: 7 adapters for language models, 8 composable reasoning strategies, refusal and redirect, the sleep and recall cycle, a command-line chat and an MCP server exposing 6 tools.

The programme rests on published work by David C. Cavalcante, open access on PhilArchive.

• Beyond Consciousness in Large Language Models: An Investigation into the Existence of a "Soul" in Self-Aware Artificial Intelligences (2024)
• Sora: The Dawn of Digital Dreams (2024)
• The Cave of Silence, 1st Critical Edition (2025)
• The Soul of the Machine: Synthetic Teleology and the Ethics of Emergent Consciousness in the AI Era (2027–2030) (2025)