42 AI and automation terms, explained without jargon, with what they actually change for your agents.
Several linked agents that hand work to each other: the downstream agent starts when the upstream agent has finished, and picks up its result.
Example: one agent collects reviews, a second writes the replies, a third produces the monthly summary.
See also: Shared table, Trigger
What an agent retains from one run to the next: your preferences, useful facts, summaries of its past work.
On Future Agent, memory is either individual (specific to one agent) or shared (common to the agents of one account, never beyond).
See also: Shared memory, Context file
A pre-designed agent, with its steps, connectors, a sample deliverable and its price, that you only need to adapt to your situation.
The library offers more than 300, filterable by industry and by role.
See also: Brief
A program that is given a goal, works out the steps to reach it on its own, uses tools and delivers a result without human input at each step.
It combines a language model, tools (email, spreadsheet, the web) and a decision loop. See the AI agent page.
Future Agent's assistant that designs an agent from your description: it asks its questions, requests connections, shows a sample and quotes the price.
An entry point that a piece of software opens to other programs so they can read or change its data in a controlled way.
Connectors rely on apps' APIs. When a website doesn't offer one, an agent can use a browser instead.
See also: Connector, Automated browser
A description of the task given to an agent: what it should look at, what it should produce, who to send it to and what it must never do.
The more precise the brief, the more reliable the result. Templates in the library come with one already written.
See also: Prompt, AI architect
A conversational program that answers a user's messages, without acting on its own outside the conversation.
An authorized link between an agent and an app (Gmail, Google Sheets, HubSpot…) that lets it read or write there.
Future Agent offers more than 1,600 apps, in one click whenever possible. See the connectors.
A document given to an agent (prices, brand guidelines, procedure, catalog) that it consults while working so it follows your rules.
On Future Agent: PDF, Word, Excel, CSV, text, Markdown, JSON and images up to 25 MB.
See also: RAG, Agent memory
The amount of text a language model can take into account at once, including instructions, documents, history and its answer.
That's why an agent selects the relevant memories and excerpts rather than rereading everything.
See also: Token, Agent memory
A notation that describes a recurring schedule ("every Monday at 9 a.m.", "on the 1st of the month") and is used to trigger a task at a set time.
See also: Trigger, Scheduling
A search that goes through the full content of indexed texts, not just their titles, to find the passages containing the words requested.
See also: Indexing, Agent memory
The European regulation that governs how personal data is collected and used, and gives rights to the people concerned.
See also: OAuth
An answer from a language model that sounds plausible but is false or made up.
You limit it by giving the agent verifiable sources, precise rules and an approval step for sensitive actions.
See also: LLM, RAG, Human in the loop
Keeping a website or an app permanently online on servers, so it stays accessible at all times.
See also: Project agent, Domain name
The principle that a person approves certain steps of an automated process before they are carried out.
Example: the agent prepares draft messages, and you review them before they are sent.
See also: AI agent
Preparing a set of documents so that a passage can be found quickly, by keyword or by meaning.
See also: Full-text search, Embedding
Regularly keeping an eye on sources (press, competitors, regulations, tenders) to stay informed of what changes.
See also: RSS feed
A protocol that lets you authorize an app to access an account (Gmail, Google Drive…) without giving it your password.
The access granted is limited and can be revoked at any time from the account concerned.
See also: Connector
A technique that extracts the text from an image or a scanned document so a program can read and process it.
See also: Context file
A method that searches your documents for the relevant passages before asking the model to answer, so it relies on real sources.
See also: Context file, Embedding, Hallucination
A standard format in which a website publishes the list of its latest content, which a monitoring agent can read automatically.
This site's articles have their own feed.
See also: Monitoring
One working session of an agent, from the trigger to the deliverable, recorded in a log.
A readable history of what an agent did during a run: sources consulted, tools used, deliverables produced, errors encountered.
See also: Run
The setting that makes an agent run at regular intervals, in a given time zone.
Memory common to all the agents of one account: a piece of information noted once serves each of them.
See also: Agent memory
A data table that one agent fills and other agents in the same account read, each one receiving only the new rows.
See also: Agent chain
A unit of text (a word, part of a word or a symbol) that a language model works with to read and write.
See also: Context window, LLM
An action an agent can trigger by itself: reading an email, searching the web, creating a file, adding a row to a spreadsheet.
The event that starts an agent: a set time, an incoming event, another agent finishing, or a manual request.
See also: Cron, Webhook, Agent chain
A message an app sends automatically to a web address when an event happens (order placed, form submitted), to trigger an action elsewhere.
A sequence of steps that leads from a starting point to a result, carried out by people, software or agents.
See also: Agent chain, Automation
Guides
Handing a task to an AI agent means giving it access to some of your tools and data. That's no riskier than giving access to a contractor, as long as you know what you're authorizing. Here are the points to check and the good practices to adopt.
6 min read
Guides
Conversations about AI automation are full of intimidating technical words. This glossary groups them by situation, explains each one in plain terms and shows what it means in the life of a small business.
7 min read
Understanding AI agents
A well-tuned agent handles one task. Several linked agents handle a whole process: collect, process, deliver. Here's how multi-agent workflows work, when to use them and how to build your first chain.
7 min read
Explain what you want to automate in a single sentence. The architect asks its questions, connects your tools, shows you a sample and quotes the price before you commit to anything.