Loading…
Loading…
Short, accurate definitions for the terms that come up when you put an AI agent into real software. Each one defines the idea on its own merits first, and only then says how it works here.
A system that uses a language model to decide which actions to take, then takes them — as opposed to only producing text.
Read the definitionAn umbrella term for AI systems that plan across multiple steps and act on the world, rather than answering one prompt at a time.
Read the definitionThe mechanism by which a language model requests that a specific function be run with specific arguments.
Read the definitionRetrieving relevant documents and supplying them to a language model so answers are grounded in specific sources.
Read the definitionAn open protocol for exposing tools, data and prompts to language models through a consistent interface.
Read the definitionA machine-readable specification describing a REST API: its endpoints, parameters, schemas and responses.
Read the definitionAn external system an agent can reach, together with the specific actions that were reviewed and approved on it.
Read the definitionThe binding between a reviewed action and a specific account on the provider, carrying a reference to a stored credential.
Read the definitionAn opaque pointer to a credential held in a vault, used in place of the credential itself.
Read the definitionTrading a token issued by one system for a scoped token issued by another, so an application can act with narrowed authority.
Read the definitionA JSON document publishing the public keys a party uses to sign tokens, so verifiers can validate signatures without shared secrets.
Read the definitionA registered identity provider whose signed assertions a platform is willing to accept.
Read the definitionA non-human identity used by software to authenticate, carrying only the permissions it was granted.
Read the definitionOne deployment serving many customers, with each customer’s data isolated from the others.
Read the definitionA database feature restricting which rows a query can see, enforced by the database rather than the application.
Read the definitionA design where a person approves or rejects a consequential action before it executes.
Read the definitionConstraints that bound what an AI system can do, distinct from instructions that ask it to behave.
Read the definitionAn attack where instructions hidden in data an AI system reads are treated as commands from its operator.
Read the definitionA durable record of what a system did, who caused it, and what it touched — sufficient to reconstruct events afterwards.
Read the definitionA property where repeating an operation has the same effect as performing it once — essential when retries are possible.
Read the definitionThe documentation covers how these ideas fit together in a system you can actually run.