Each word gets a short meaning. Where a page says more, there is a link.
#A–B
Actor — who an action runs as. That can be you. It can be one of your automations. It can be an outside party with a contract, such as a connected AI agent or a customer. Every audit entry names its actor.
Approval — an action that is paused and waits for your decision. It keeps waiting until you answer. Often you can change its details first. See Grants and approvals.
Archive — one encrypted file that ends in .recued.archive. It holds a
full backup of your server. See
Backup and restore.
Audit — the diary. Every run writes a record on your server. It says who asked (the actor), on which channel, what was read, what was written, and how it ended. See Audit and memory.
Batch mode — one AI call that runs a contracted AI function on every item
in a list. You turn it on with llm.id_field.
Browser Bridge — a small add-on for your web browser. It watches for the page triggers a recipe declared. It reads web pages and does things on them, like clicking or typing. It shows notifications from your computer's own system. It does all of this only for your server. It runs only signed commands from the server. It holds no data.
BYOK — bring your own key. These are your paid or self-hosted AI models. They go in the two slots for normal AI. See Set up AI.
#C
Category — a seller's own label that groups access tiers with their
customers. Every server starts with one, called main, and most sellers
never change it. A second category is only for selling separate products
whose tier keys would clash. A category cannot change once a tier exists.
Circuit breaker — a safety switch. When a reactive recipe fails many times in a row, the switch turns it off and tells you. You turn it back on yourself.
Claim link — a link sent to a customer by mail. It does one job. The customer follows it to pick up their access token. The token itself is never sent by mail. See Selling access.
Connection — a saved way to talk to an outside service. It has a name, and you can use it again and again. The service can be an API, an MCP server, or a place to send notifications. The secret keys stay on your server. See Connections.
Contract — the saved list of what someone may do. The gate that checks every action reads it. It can cover an installed capability, a chat session, or a customer tier.
Customer — an outside person or company that a seller gives access to, at a chosen tier. A customer connects with any MCP client or any OpenAI-compatible chat client. They do not need Recued.
#D–H
Door — a way in from outside that you open on purpose. There are MCP doors, reception doors, and messenger doors. Each door has its own grants, tool by tool, and its own secret key.
Enforcement boundary — the one gate every action goes through. The gate works out who is asking and which capability they use. It checks the grant and the risk tier. It runs the action on your server. Then it writes the audit record.
Embeddings slot — the one AI provider and model used for work that makes vectors. Some recipes and some housekeeping need that. Chat never uses it. Normal AI never uses it. If it is missing, nothing falls back to the free pool or the BYOK slots.
Enrichment — a fact Recued works out ahead of time and sticks onto a warehouse record. A score or a signal, for example. Some come from plain math. Some come from AI, under the trust level you set for each topic.
Free pool — your own free-tier AI keys. Recued tries these before your paid keys.
Grant — a yes you give ahead of time. It says which actions a pack, a door, or a session may use. It says at what access level. And it says who may use them.
Housekeeping — clean-up work the server does while the server is not busy. It packs the database, clears old cache, refreshes enrichments, and checks CRM data against the vendor.
#I–M
Ingredient — one action a recipe can use. A small file (a manifest) describes it. It can be an HTTP call, a page action, an AI function, a storage call, or a connection call. The file says what goes in, what comes out, and how risky it is. See Ingredient manifest.
Kitchen — the place where you make, copy (fork), and test recipes and ingredients before you publish them.
LLM gateway — your server's chat endpoint that speaks the OpenAI shape.
It answers at /v1/models and /v1/chat/completions. Any chat app that
takes a base URL and an API key can use it. Each turn follows the rules of
the contract it was granted. See
Connect to Recued.
Marketplace — the public store of recipes, ingredients, and packs. Looking and installing are free. Publishing needs a free account.
MCP — an open, shared language that outside AI apps use to call your server's tools. See AI chat and MCP.
Memory — your own notes: facts, choices, and things you let an AI save. You own it. Nothing prunes it. Recall searches it. It is not the audit diary. See Audit and memory.
#P
Pack — a box that holds recipes, ingredients, and rules. You install the whole box at once. Grants attach to the pack. See Recipes, ingredients, and packs.
Pairing — linking a device (the webclient or the Bridge) to your server. The device gets a limited pass. You can cancel it. Pairing goes straight to your server, never through the cloud.
Pro — a paid extra, $5.99 USD a month, and you can skip it. It gives your
server a <handle>.recued.net name and a certificate that renews by itself.
It unlocks no features.
Publisher handle — your name in the marketplace. It is free. Everything you publish lives under it.
#R–S
Reception — your server's public front door. It can show pages, forms, booking links, and file drops. Every request that comes in waits for you to review it. See Reception.
Recipe — a list of steps written in JSON that you can read. The steps use ingredients and transforms to do one job. See Authoring recipes.
Recovery key — 24 secret words. They let a device pair. They unlock the data your server encrypts. They also open your backup archives.
Server — the Recued program you run on a machine you control. It does all the real work. It is the one part everything else trusts. It holds the warehouse and your secret keys. It runs the timers. It owns the audit diary. It is the only part that touches your accounts.
#T–W
Tier — a named level of access a seller offers. Behind each tier is a customer contract template. That template fixes what the tier grants and how much it may be used.
Timeline — one feed, in time order, for one thing, such as a person or a deal. It is built from provenance links across mail, calendar, CRM, notes (annotations), and memory.
Transform — a small safe tool a recipe uses to work on data. It always gives the same answer for the same input. It costs nothing. See Transforms.
Trigger — the web page patterns (URLs) a recipe is made for, so it can
auto-run there. This is not the same as trigger_steps. Those are the check
steps a reactive recipe runs to decide whether to go ahead.
Warehouse — the place on your server that keeps a copy of your mail, calendar, contacts, files, and CRM records. Recipes and AI answers are based on it. See The warehouse.
Watcher — the trigger part of a reactive recipe. It watches mail, files, the calendar, the clock, or web pages. Each tick, it decides whether the recipe should go ahead.
Webclient — the app in your web browser at app.recued.com. Your own
server can serve it too. It shows things and takes your clicks and typing.
It does no
real work and holds no secret keys.