Skip to main content

Key concepts

This page is a small glossary. Each term is explained in one or two sentences, with a link to the page that covers it in depth. If a word in the rest of the documentation is unclear, it is probably here.

Using the agent​

TermWhat it means
AgentThe AI assistant. Unlike a plain chatbot, it can call tools to fetch information or perform actions, then answer using the results.
ProfileA ready-made "version" of the agent for a job: its instructions, the tools it can use and, optionally, the AI model. Examples: aion_std (general assistant), file_server_assistant, infra_sre. You pick it in the chat; in the API it is the profile field (or model in the OpenAI-compatible API). → Profiles
SkillA short written procedure the agent can read when needed (for example "how to build a monthly report"). Skills teach the agent how to do something; tools give it the means. → Skills and prompts
ToolA single action the agent can perform, such as reading a file, searching the web or querying Prometheus. In the chat you can see which tools were used for each answer.
Connector (MCP server)A plug-in that provides a group of tools for one system (file server, Microsoft 365, a database…). MCP (Model Context Protocol) is the open standard used to plug them in. Your developers can also build their own. → MCP registry, Custom MCP servers
Conversation / sessionOne chat thread. It has an id, its history and its own workspace: a private folder with the files you upload and the files the agent creates. In the API, conversation_id and session_id are the same thing.
TurnOne question from you plus the complete answer from the agent, including all the tool calls in between.
Thinking (reasoning)Some AI models "think" before answering. In the chat you can choose how much (off, minimal, medium, maximum): more thinking is slower but better for hard problems.
StreamingThe answer appears word by word while it is generated, instead of all at the end. The API offers both modes.

Memory​

TermWhat it means
Short-term memory (STM)The recent messages of the current conversation that are sent to the AI model at each turn. When the conversation gets long, older parts are summarized (compaction) so it fits. → Context compaction
Long-term memory (LTM)Facts the agent learns about you across conversations (preferences, recurring information) and reuses later. → Memory overview
Instructions (USER.md)Personal instructions you write in the chat settings ("answer in Italian", "I work in the finance team"), added to every conversation. → SOUL, MEMORY and USER
Query memory / projectA shared catalogue of validated database queries, grouped by project (for example sales), so the agent reuses queries that are known to be correct.

Access and security​

TermWhat it means
UserA person with an account. Can sign in with a password or, if enabled, with a Microsoft or Google company account (SSO).
AdministratorA user with the admin role: can open the admin panel and manage users, models, profiles and connectors.
API keyA secret string (aion_…) that lets a program call AION on behalf of one user, with limited permissions (scopes). → Get an API key
ScopeA permission attached to an API key: chat, conversations:read, conversations:write, files:read, files:write.
Service accountA non-human account (svc-…) for integrations that should not depend on a real person. Only administrators create it and its keys. → Service accounts
Personal credentialsUsernames, passwords or tokens that you enter once in Integrations so a connector can act as you (for example your Windows account for the file server). They are stored encrypted.

Running AION​

TermWhat it means
LLM providerThe service that gives AION access to an AI model. AION works with the model your company chooses; the administrator configures it. → LLM provider precedence
chat-ui / admin-ui / backendThe three main applications: the chat web app, the admin web app and the API server that does the work.
Tenant / instanceOne AION installation for one company, usually reachable on its own domain.