Docs / Reference
Glossary
Every term Mindset uses, in plain English, with the invoice exceptions workflow as the example where one helps.
Agent
The thing that does the work. It holds a system prompt, a script, a set of resources it may reach, and a model. It is what gets triggered and scheduled.
Example: The invoice exceptions agent works through four stages and produces a supplier query for approval.
Not: A chatbot with tools bolted on, and not a workflow you drew. You write the outcomes and it finds its own way to them.
BAC (behaviour acceptance criterion)
A situation you put the agent in, plus a statement of what must be true about how it behaved. Pinned criteria control releases: a new version goes live only if every one passes.
Example: "Given an invoice with a £40 difference, the agent calls the invoice-comparison function."
Not: A unit test of your code, and not editable once pinned.
Bundle
Everything an agent carries: its prompt, script, model and resources. Giving somebody the agent gives them the bundle, exercised through that agent.
Example: Publishing the invoice agent gives the holder read access to your finance system through four operations, plus the ability to register supplier queries for approval.
Not: Access to the systems themselves. Only through the named operations, only in the order the script allows.
Check
A straight test of what a stage collected. Fast, free, and not a matter of opinion.
Example: Stage one passes when both the invoice and its purchase order have been retrieved.
Not: A model's opinion. That is a judgement.
Connection
A link to one outside system, with the login details for it. Those details are held by us. They never reach the agent and never reach the model.
Example: The finance system, and the supplier contracts knowledge base.
Not: What you give an agent. You give it operations.
Environment
A named partition inside your org: Demo, Test, Production. Any member can create one, list them and switch. Nobody can rename, archive or delete one. Every org starts with one called original.
Not: A copy of your work. Changes do not travel between environments on their own.
Function
A fixed list of steps that gives the same answer for the same input. An agent calls it as a tool. A step calls an operation, calls a model, or works on data it already has.
Example: The invoice-comparison function: fetch the PO, extract PO lines, extract invoice lines, match and list differences, return the differences plus their total value plus whether there are any at all.
Not: Something you trigger or schedule. That moved to the agent. A function can still call a model inside a step.
Hub
The colleague-facing conversation surface, where people use the agents published to them.
Example: Somebody in finance opens the Hub and asks the invoice agent about INV-4471.
Not: The place you build. That is the Agents workspace.
Judgement
Another model reads what a stage produced and decides whether it is good enough. Only a clear yes passes, and the reasoning is recorded whether it passed or failed.
Example: Stage three passes only if the answer names a specific contract clause rather than referring to the contract in general.
Not: The first choice. Use a check wherever the answer is yes or no.
Operation
One specific named thing an agent can do on a connection, with a fixed input, a fixed output, and marked as either read or write. Operations are discovered rather than hand-written.
Example: "Get the purchase order for an invoice number" (read) and "register a supplier query" (write).
Not: Switched on by itself. Nothing enables without you enabling it, and enabling makes one real call to prove it works.
Orca
The provisioning conductor, short for Orchestrator. You describe what you want, it looks at what you already have, proposes a plan as a card, and stops. It authors nothing itself: each piece is built by a specialist builder.
Not: Where you edit something that already exists. That is the Agent Builder, docked on the left of every agent tab.
Org
Your workspace, and the only isolation boundary. Nothing sits below it. Its region is fixed when it is created.
Run
One execution of an agent, from whatever started it to however it ended. Every run is recorded: the trigger, the stages, the operations called, what came back, and any runs it delegated to other agents.
Not: Resumable. A stopped run does not continue, and re-running starts from the beginning.
Script
The ordered stages an agent works through. Each stage says what to tell the agent, what it is allowed to use, what it must collect, and the condition for moving on. Routes only go forward.
Example: Gather, Compare, Assess, Prepare.
Not: A flowchart of steps. Nobody writes the steps.
Stage
One step of a script, defined by an outcome rather than an action. The agent works out how to reach it; Mindset decides when it has.
Example: Stage two must have the line-by-line differences between the invoice and the PO.
Not: Something the agent can declare finished on its own.
System prompt
What the agent is and how it behaves, all the time, on every stage.
Example: "You handle invoice exceptions for the finance team. You are precise about numbers and you never guess a contract term."
Not: The place to put a rule that needs a definition. "A material discrepancy" belongs in a function as a threshold, not in wording.
Widget
Something an agent can put in front of a person during a conversation instead of answering in plain text. Versioned like a function: save creates a version, publishing checks it will render, and rollback is re-activating an earlier one.
Write
An operation that changes something in an outside system. A write does not happen when the agent calls it. The agent records exactly what it would send, the run carries on and finishes, and the change waits for a person to open a link and approve it.
Example: Registering a supplier query. The supplier hears nothing until somebody in finance approves it.
Not: Something an agent can approve, its own or another agent's. Only an org policy in Settings → Governance changes that, and then the record names the policy.