User-facingOracle
Oracle — system instruction + RAG context
Prepends retrieved node context after this block. The model answers only from those nodes. Voice-tier instructions are injected between the persona and context blocks.
The Elephant Observatory
AI transparency
System prompts & instructions · as deployed
What we tell the models before your messages arrive.
Why we publish this
A system prompt is the instruction block a model sees before it reads your question or feedback. It sets tone, constraints, and role: what the model is allowed to do, what it should refuse, and how it should cite or stay within TEO's knowledge base.
We publish these texts so you can judge the product on its merits — not only the glossy surface. If something feels off in the Oracle or Hermes, you can read the same instructions we ship.
Privacy in plain terms: this page shows our side of the contract — templates and defaults. Your messages to Hermes or the Oracle are not listed here; they are processed according to our Privacy Policy and the terms linked in the footer.
What to look for: grounding and source-first rules (e.g. answer only from retrieved nodes), instructions that resist flattery or false certainty, safety boundaries, and the same structural honesty as our BYOD stance. Placeholder text substituted with live data in production is noted inline where relevant.
Prompt catalog
Each card shows a representative version of the prompt. Some entries use sample data where the real call is dynamic. Use the tabs below to browse by system.
User-facingOracle
Oracle — Apprentice voice instruction
Injected when the user selects Apprentice voice. Uses description_apprentice from each node and this tone directive.
User-facingOracle
Oracle — Magus voice instruction
Injected when the user selects Magus voice. Uses description_magus from each node and this tone directive.