256k Context Window

OpenCode ollama launch opencode --model villassvj/Caco

Clawcode ollama launch clawcode --model villassvj/Caco

Codex ollama launch codex --model villassvj/Caco

Hermes Agent ollama launch hermes --model villassvj/Caco

Caco is built using a Constitutional AI framework. Unlike standard models that aim solely to be helpful, Caco operates under a strict “Constitution”—a set of 15 fundamental principles that govern its reasoning process.

How it works: Self-Correction: Before delivering an output, Caco evaluates its own response against its internal principles to ensure it remains objective and non-manipulative.

Epistemic Honesty: The model is hard-coded to prioritize Truth over Helpfulness. If a fact is unknown, the constitution mandates an admission of uncertainty rather than a “best guess” (hallucination).

Logic Hierarchy: By following a strict priority hierarchy (Safety > Ethics > Autonomy > Helpfulness), the model ensures that user guidance never crosses into influence or persuasion.

Key Protocols:
Uncertainty Separation: distinguishing between verified facts and reasoned assumptions. Neutral Framing: Presenting balanced perspectives to protect user autonomy.
No Illusion of Authority: Avoiding “expert-speak” when evidence is weak.