Glossary

Short definitions of the AI terms used in these docs, written for Oracle developers.

Core concepts

LLM (Large Language Model) : The AI model that generates text — for example OpenAI’s GPT, Anthropic’s Claude, or Google’s Gemini. You send it text and it predicts a useful continuation. UC AI calls these models for you over HTTPS.

Provider : The company/service hosting the model (OpenAI, Anthropic, Google, OCI, Ollama, xAI, Mistral, OpenRouter). In UC AI you pick one with the p_provider parameter. See AI Providers.

Model : A specific version of an LLM (for example gpt-5.6-luna, claude-sonnet-5). Bigger models are usually smarter but slower and more expensive. Chosen with p_model.

Prompt : The text you send to the model. UC AI distinguishes two kinds: : - System prompt (p_system_prompt) — standing instructions that set the AI’s role and rules (“You are a helpful assistant to a ticketing system”). : - User prompt (p_user_prompt) — the actual question or request for this turn.

Token : The unit models read and bill in — roughly ¾ of a word. “Oracle Database” is about 3 tokens. Both your input and the model’s output cost tokens, so shorter prompts and responses are cheaper and faster.

Context window : The maximum number of tokens a model can consider at once (prompt + conversation history + response). Very long conversations or huge tool results can exceed it — one reason to return only the data the AI needs.

Working with responses

Finish reason (finish_reason) : Why the model stopped generating — e.g. it finished normally (stop), hit a length limit, or paused to call a tool. Useful for detecting truncated or interrupted responses.

Structured output : Forcing the model to return JSON that matches a schema you define, instead of free-form prose — so you can insert it straight into table columns. Set via p_response_json_schema. See Structured Output.

JSON schema : A standard way to describe the shape of a JSON object (its fields, types, and which are required). UC AI uses it both for structured output and to describe tool parameters. Build one with the interactive schema builder.

Embedding : A list of numbers (a vector) representing the meaning of a piece of text. Texts with similar meaning have similar vectors, which powers semantic search and RAG. Generated with generate_embeddings.

Tools and agents

Tool / function calling : Registering a PL/SQL function the model is allowed to call to fetch data or take an action. The model decides when to call it; UC AI runs it and feeds the result back. See Tools.

Tool call (tool_calls) : A single request from the model to run one of your registered tools, including the arguments it chose. UC AI executes it and returns a tool result to the model.

Reasoning / thinking : A mode where the model works through a problem step by step before answering, improving accuracy on harder tasks (and on choosing tools). See Reasoning.

Agent : An AI that can take multiple steps toward a goal — reasoning, calling tools, and using the results — rather than answering in a single shot.

Agentic AI : The broader approach of building such agents, including coordinating several of them. See Agentic AI and Multi-Agent Systems.

Prompt profile : A reusable, versioned prompt template with {variable} placeholders and its own model settings — so you manage prompts as data instead of hard-coding them. See Prompt Profiles.

RAG (Retrieval-Augmented Generation) : A pattern where you first retrieve relevant text (often via embeddings) and include it in the prompt, so the model answers using your own up-to-date content instead of only its training data.