Glossary

Key Opal terms and concepts, listed alphabetically for quick reference.ne in more depth.

• Activity log — The record of the actions and updates made to a resource, showing what changed and who changed it. The organization-wide view of this record is the audit log.
• Agent — A digital worker in Opal. Agents collaborate alongside people: taking part in conversations, completing and reviewing tasks, executing and reviewing steps in workflows, drawing on organizational knowledge, and interacting with connected systems.
• Agent template — A reusable blueprint for creating agents, capturing a pre-configured prompt, model, and capabilities so new agents start from a ready-made configuration instead of a blank one. Agent templates are organization-wide, and every agent created from one stays linked to it.
• Agentic operations — The practice of coordinating people, agents, knowledge, systems, and processes toward shared business objectives. It is the operating model Opal is built around, and the product category Opal belongs to.
• AI Assist — Assisted authoring in the Widgets section: describe the widget you want in plain language and Opal generates its editable code and schema automatically, ready to be refined.
• Allow Dynamic Step Creation — An optional behavior on a flow step that lets it add further steps during execution when the work calls for it. Steps added this way are added linearly, in keeping with a flow's sequential execution.
• Archiving — Retiring a file or folder in Knowledge so it is no longer part of the working knowledge base. It is one of the management actions available on knowledge content, alongside renaming and moving.
• Assignee — The agent or person a flow step is assigned to, who carries out the work described in that step's instructions.
• Auto-discovery — What a verified email domain enables: people whose email address matches a domain the organization has added and verified can join, or request to join, that organization based on that address, rather than waiting for a direct invitation.
• Automatic top-up — A purchase of credits made automatically whenever an organization's credit balance falls below a threshold it has set. Automatic top-ups are configured in Organization Settings, alongside one-time top-ups and recurring credit subscriptions.
• Billing alert — A notification that keeps the people responsible for an organization's funding informed about its billing and credits, so a change in the balance surfaces without anyone having to check. Billing alerts are configured in Organization Settings.
• Billing profile — An organization's billing details, held once at organization level. The billing profile is managed in Organization Settings, alongside the organization's payment methods, subscription plan, and seats.
• Child team — A team nested inside another team, used to represent a function within a larger group — for example a team for a department, with child teams for the specific functions inside it. Child teams are created and managed with the rest of an organization's teams in Organization Settings.
• Component library — The set of interface components available in the Widgets section, which widgets are assembled from.
• Configuration field — An optional field on a tool definition that is shown to whoever creates a tool from it. Configuration fields capture values that stay the same each time the tool runs but differ from one tool to the next — for example a base URL, an account or workspace ID, or an API key.
• Connector — A secure link between Opal and an external tool or system, built on the Model Context Protocol (MCP), that an agent can use to retrieve data and take action in that system.
• Continuous Learning — An optional behavior on tasks and flows that captures observations and lessons from each successful run so the work improves over time. In flows, those lessons are compiled into a Continuous Learning document that can be viewed and edited.
• Credit transaction — An entry in the record of movements in an organization's credit balance. Credit transactions are viewable alongside the balance in Organization Settings.
• Credits — The balance an organization uses to fund its usage of Opal. Credits can be topped up on demand, purchased on a recurring subscription, or added automatically when the balance falls below a set threshold.
• Custom role — A role an organization defines for itself, for use when Opal's platform-wide preset roles do not match the way it works. Custom roles are assigned to individual members or to teams, in the same way preset roles are.
• Detection method — How a guardrail identifies the content it is looking for. A guardrail uses one of three: keyword matching for exact words or phrases, a regex pattern for more complex patterns, or semantic analysis for meaning and context.
• Digital worker — A worker that is software rather than a person, able to reason, decide, retrieve information, and carry out work autonomously or semi-autonomously. In Opal, digital workers are called agents.
• Digital workforce — The combination of human and digital workers an organization coordinates together. People remain responsible for leadership, judgment, creativity, and governance; digital workers add capacity for analysis, coordination, and execution.
• Draft — The working copy in which changes to a resource are made before they go live. Publishing a draft creates a new version; until then, the published version remains what everyone else works from.
• Email domain — A company email domain an organization adds and verifies in Organization Settings. Once verified, it enables auto-discovery, so people whose email address matches it can join, or request to join, the organization.
• Embedded artifact — Content contained inside another file, such as an image within a document. File processing can extract an embedded artifact into its own file in Knowledge.
• Execution instructions — The instructions an agent follows when it is assigned to carry out a task or a step in a flow. When an agent is assigned to review one instead, it follows that assignment's review instructions.
• Executor–reviewer pattern — Assigning one agent to carry out a piece of work and another to review it before the output is accepted, so quality is checked without a person inspecting every result.
• Fallback note — A note a template author writes for something included in a template — a skill, connector, tool definition, widget, guardrail, attached knowledge, or a default agent — shown to anyone who does not have access to it.
• File processing — The file-type-aware handling Opal offers for content added to Knowledge, which turns a raw file into something people and agents can use — for example digitizing a scanned PDF into markdown, transcribing audio or video with speaker diarization, annotating an image or video, or extracting an embedded artifact into its own file.
• Flow — A multi-step workflow that coordinates people and agents through a complex process from start to finish. Each step can be assigned to an agent or a person, with optional review at every step.
• Flow-level context — The overarching goals, requirements, and purpose set for a whole flow, so every step runs with the same understanding of the objective. Additional context can also be supplied for an individual run.
• Flow template — A reusable blueprint for creating flows, capturing a pre-configured flow — its step sequence, step instructions, assignments, reviewers, attached resources, and flow-level context — so new flows start from a ready-made, multi-step setup instead of an empty workflow. Every flow created from one stays linked to it.
• Folder — A grouping of related files within Knowledge's centralized, file-system-like structure. Folders can be renamed, moved, and archived as the knowledge base changes.
• Governance — The oversight and control built across the Opal platform, including permissions, approvals, versioning, auditability, and safeguards on how agents behave.
• Grounding context — The organizational knowledge an agent draws on to inform its work. Knowledge becomes grounding context when files and folders are added to a space: the agents operating there retrieve and ingest that content while working in threads, tasks, and flows.
• Guardrail — A safeguard applied to what goes into an agent and what comes out of it. Each guardrail is attached as an input or output guardrail and defines how flagged content is handled — for example, redacting, hashing, masking, or blocking it.
• Handling approach — What a guardrail does with the content it flags: redact it with a placeholder, mask its characters, replace it with a cryptographic hash, or block the entire input or output and display a specified error message. Each guardrail carries a recommended approach, which can be changed for each place the guardrail is attached.
• HTTP tool — A tool whose definition carries out its action by calling an external service over HTTP, using a URL, an HTTP method such as GET or POST, and any headers the service requires (header values can use variables). An HTTP tool calls an existing endpoint; it does not create one.
• Input guardrail — A guardrail attached so that it validates user input before an agent processes it.
• Input modality — A kind of input a model can accept natively, such as text, images, audio, or video. When an agent retrieves a file from Knowledge, Opal passes the file directly if the model supports its format as a native input modality, and supplies a text equivalent — such as a transcription — if it does not.
• Invitation — The direct route into an organization: an administrator invites a person to become a member, rather than waiting for them to discover the organization through a verified email domain.
• Join request — A request to become a member of an organization, made by someone who has discovered it through a verified email domain. Administrators review and approve join requests, so membership stays under administrative control.
• Knowledge — Opal's centralized knowledge base: a single, organized home for the documents, files, and shared context an organization relies on, structured as a familiar system of files and folders. It gives people and agents a common source of truth to work from.
• Library — A standalone grouping of files and folders in Knowledge, used to bring related content together for a particular purpose.
• Linked resources — The connections Opal tracks between resources, showing where a given resource is being used — for example, which agents use a model config — so the impact of a change can be understood before it is made.
• Listing — A resource an organization has published to the Marketplace for other organizations to adopt, either for a price or for free. A listing can be submitted for verification and can be taken down at any time.
• Marketplace — Where organizations list the resources they have built for other organizations to use, and find and adopt resources built by others. Resources can be listed for a price or for free.
• Marketplace profile — An organization's public-facing presence on the Marketplace. An organization sets up a marketplace profile and configures its payout settings before it can create listings and sell; the profile is managed at organization level, alongside its listings and payouts.
• MCP Connection — One person's link to an MCP Server. Each user creates their own, so a connection carries that individual's permissions and available functionality in the connected system. Connections can be attached to agents, spaces, tasks, and flows.
• MCP Server — The definition of an external service Opal can connect to, holding the service's URL and its authentication type — None, Bearer Token, or OAuth. Users create MCP Connections from a server, and a server can also be attached to agent, task, and flow templates as a definition.
• Member — A person who belongs to an organization on Opal. Members are invited directly by an administrator or join through a verified email domain, and are then organized into teams and granted access through roles.
• Model — The underlying AI model that powers an agent's reasoning. A model can be selected directly with its parameter values set, or selected as a pre-defined model config.
• Model config — A reusable preset that pairs a specific model with a defined set of parameter values, so the same configured model can be applied when creating agents and agent templates.
• Model Context Protocol (MCP) — The protocol Opal's connectors are built on for integrating with external tools and systems. Connected services must support remote MCP; local MCP servers are not supported, as Opal is a cloud-based platform.
• Model parameters — The settings that shape how a model behaves. A model's parameters are shown alongside its specifications and capabilities in the models catalog, and a model config saves a chosen set of parameter values so they can be reused.
• Models catalog — The catalog of language models available in Opal, searchable by name and filterable by provider and by the input and output modalities a model supports, with each model's token costs shown in Opal credits.
• Multi-factor authentication (MFA) — Authentication that requires more than one factor to sign in, rather than a password alone. An administrator can enforce MFA for all members of an organization; it is an organization-wide security setting, configured in Organization Settings.
• OCR (optical character recognition) — The file processing option that digitizes a scanned document into markdown text, so its content can be read and searched instead of remaining locked in an image.
• Organization — A single tenant on the Opal platform: the top-level container that represents a company and holds its members, teams, roles, resources, and organization-wide settings.
• Organization preferences — The organization-wide settings an administrator configures for an organization, including its notification settings. They are set once, at organization level, rather than by each member.
• Organization profile — The identity details of an organization inside Opal: its name and its logo. It is managed in Organization Settings, and is separate from the organization's public-facing marketplace profile.
• Organization Settings — The central area where an organization's settings are managed, covering its profile and preferences, security, membership, teams, roles, billing, and credits.
• Output guardrail — A guardrail attached so that it validates an agent's output before it is returned.
• Output modality — A kind of output a model can produce, such as text or images. The models catalog can be filtered by the output modalities a model supports, alongside its input modalities.
• Participant — A person or agent that has been added to a thread and can take part in the conversation. Participation comes from being added to the thread, not from being tagged in it.
• Payment method — How an organization pays for its Opal subscription. Payment methods are managed at organization level, in Organization Settings, alongside the billing profile, subscription plan, and seats.
• Payout — The payment a selling organization receives for its Marketplace sales. Buyers are charged in credits, and sellers are paid out in US dollars, after Opal's service fee.
• Payout settings — The configuration that determines how an organization is paid for its Marketplace sales. An organization configures its payout settings, alongside setting up its marketplace profile, before it can create listings and sell; payout settings are managed at organization level.
• Permission level — The level of access someone is granted to a specific resource: when a resource is shared with a member or a team, each is granted a permission level that determines what they can do with it, and the levels available differ by resource type. For a space, the levels are Manage to view, edit, and use the space as well as manage its settings and sharing; Access to view, edit, and use it; View to see its details.
• Placeholder — A {{ placeholder }} written into a template for a detail that changes from one item created from it to the next. In an agent template, placeholders are written into the prompt and each carries an author-written description that tells whoever creates the agent what to enter. In a task template, they are written into the task instructions and the agent reviewer instructions, and are filled in by whoever creates the task. In a flow template, they are written into the flow context, the step instructions, and the agent reviewer instructions, and are filled in by whoever creates the flow.
• Platform event — Something that happens on the Opal platform that an organization can subscribe to with a webhook, so that Opal sends it to an external system. Webhooks and their subscribed events are managed in Organization Settings.
• Preset role — A platform-wide role Opal provides, available to every organization without configuration and assigned to individual members or to teams.
• Pricing unit — The unit of usage a Marketplace listing is priced against, set by the type of resource being listed — for example per 1M output tokens for an agent template, per 1K invocations for a guardrail, or per 1K renders for a widget. The seller sets the price, or makes the resource free; buyers pay in credits according to their usage.
• Primary skill document — The main document within a skill, describing how that kind of work is performed. It is written and edited in a rich editor, and any number of reference documents can be added alongside it.
• Prompt — The instructions that define an agent's role and how it should behave. Every agent is configured with one.
• Protected prompt — A template prompt the author has hidden, so someone creating an agent from that template can supply values for its placeholders without seeing the prompt itself. A template's language model parameters can be protected in the same way.
• Publishing workflow — An optional governance control that requires a new version of a resource to be reviewed and approved by a member with the required permissions before it can go live.
• Recommended trigger — A trigger type a task template specifies — manual, recurring, schedule, or webhook — to guide how tasks created from that template should be run.
• Recurring credit subscription — A subscription that purchases credits for an organization on a monthly, quarterly, or annual cadence, funding a steady level of usage without anyone acting each time. It is separate from the organization's subscription plan, and is set up in Organization Settings.
• Reference document — A supporting document added to a skill alongside its primary skill document — for example additional documentation, a template, or a worked example — which an agent pulls in as needed when it applies the skill.
• Request Input — An optional behavior on tasks and flow steps that lets an agent pause during execution to ask for information it needs to complete the work.
• Request log — The record of the requests a webhook has sent, kept with the webhook in Organization Settings. It shows what was sent, and an individual request can be re-attempted from it.
• Reset — Returning a flow run to a state where it can be run again, usually with new context added. On the re-run, Opal analyzes the work already completed at each step and determines whether that step needs to be redone or can be skipped. A run can be paused or reset at any point.
• Resource — Something built and managed in Opal that the platform's governance controls apply to — for example an agent, skill, flow, template, guardrail, tool, or widget.
• Resource tag — An organization-defined label that can be applied to the resource types the organization specifies, used to classify resources and to filter to all resources carrying that tag.
• Reviewer — An agent or a person assigned to check work before it is accepted, following its own review instructions. A task can have an agent reviewer and a user reviewer, as can any step in a flow; reviewing a flow step means approving it, with optional comments, or requesting changes with written feedback.
• Role — What an administrator assigns to a member or a team to determine what they can do across the organization. Opal provides platform-wide preset roles, and an organization can also define custom roles of its own.
• Role-based access control (RBAC) — How Opal governs what people can do across an organization: administrators assign platform-wide preset roles or custom roles to individual members or to teams.
• Run — A single execution of a task or a flow. A task's run is carried out by its assigned agent, following the task's instructions, and every run is recorded in the Run Log.
• Run Log — A record of every execution of a task or flow, including the actions taken, so results are traceable and easy to audit.
• SAML single sign-on (SSO) — An organization security setting, available on the Enterprise plan only, that lets members authenticate through the company's own identity provider rather than signing in to Opal separately. It is configured in Organization Settings.
• Seat — A place on an organization's subscription. Seats are managed as part of billing, in Organization Settings, alongside the billing profile, payment methods, and subscription plan.
• Security settings — The organization-wide controls over how people are admitted to an organization and how they authenticate: enforced multi-factor authentication for all members, SAML single sign-on on the Enterprise plan, and the email domains the organization has added and verified. They are configured in Organization Settings.
• Semantic analysis — The guardrail detection method that uses a lightweight language model to detect meaning and context rather than exact matches — for example, detecting PII or content that violates a company policy.
• Service fee — The share of each Marketplace transaction that Opal retains: 20%, with the remaining 80% paid out to the selling organization.
• Skill — Reusable, packaged expertise — instructions and procedures — that guides how an agent performs a specific kind of work. A skill is made up of a primary skill document and any number of reference documents, and can be attached to agents, spaces, tasks, flows, and templates.
• Space — An operational environment where people and agents come together to do work. A space brings the agents, knowledge, capabilities, and work of a team, project, or initiative into one organized context.
• Speaker diarization — The part of audio and video transcription that distinguishes between speakers, so a transcript shows who said what.
• Step — A single unit of work within a flow, made up of a name, an optional description, an assignee, and instructions. Any step can also have an agent reviewer, a user reviewer, or both. A flow's steps execute sequentially, one at a time.
• Subscription plan — What an organization is subscribed to on Opal. The plan is managed in Organization Settings with the rest of billing — the billing profile, payment methods, and seats — and it is one of the gates on what the organization can use: some capabilities, such as SAML single sign-on on the Enterprise plan, are available only on particular plans.
• Sync — Bringing something created from a template up to date with the template's latest updates. An agent, task, or flow created from a template stays linked to it but never changes automatically; the update is applied only when a user chooses to sync.
• Tagging — Using @ in a thread message to invoke an agent that is already in the thread to respond, or to notify another user and draw them into the conversation. When several agents are tagged in one message, they respond in the order they were tagged.
• Task — A unit of work assigned to an agent to run on its own, without a person directing it in a conversation. A task carries its own instructions and can be triggered on demand, on a recurring cadence, at a scheduled time, or by an external event.
• Task template — A reusable blueprint for creating tasks, capturing a pre-configured task — its instructions, assigned agent, reviewers, recommended trigger, and attached resources — so new tasks start from a ready-made setup instead of a blank one. Every task created from one stays linked to it.
• Team — A group of members within an organization, used to organize people and to assign access. Teams can contain nested child teams, and roles can be assigned to a team as well as to an individual member.
• Template — A reusable, pre-configured starting point for creating an agent, task, or flow, so proven work can be standardized and used again across the organization.
• Templated field — A field in a template written with {{ placeholders }} for whoever uses the template to fill in. In a task template, the task instructions and the agent reviewer instructions can be templated fields, and either can also be protected; in a flow template, the flow context, the step instructions, and the agent reviewer instructions can be templated fields, and any of the three can also be protected. User reviewer instructions are not templated and cannot be protected, so a human reviewer can always read them.
• Thread — A conversation between people and agents inside a space, where they work together in real time by asking questions and directing work back and forth.
• Token caching — An optimization Opal applies automatically to reduce token costs by making greater use of cached input and output tokens. It requires no configuration.
• Tool — A single, callable action an agent can invoke to get work done, such as sending a request to an external service. Every tool is created from a tool definition, and tools can be attached to agents, spaces, tasks, and flows.
• Tool definition — The reusable blueprint a tool is created from, describing a single action: the function's name, a description of what it does, and its parameters, plus any configuration fields. A definition can also be attached to agent, task, and flow templates.
• Top-up — A one-time purchase of credits, made when an organization wants to add to its balance on demand rather than on a cadence. Recurring credit subscriptions and automatic top-ups add credits without a purchase being made each time.
• Trigger — How a task starts: manually, on a recurring cadence, at a scheduled time, or in response to an external event through an incoming webhook.
• Variable — A token, credential, or configuration value — such as an API key or an environment variable — that an organization stores once and references when configuring other resources, including MCP Connections, Tool Definitions, and Webhooks. Each variable has a key used to reference it and a type of either text or secret, which determines how the value is handled.
• Verification — The Opal team's review of a Marketplace listing, requested by the seller. The resource is reviewed for safety and against Opal's Marketplace policies, and the verification is then approved or rejected. A verified resource carries a trust signal that helps buyers choose with confidence between listings.
• Version history — The record of a resource's published versions. Changes are made in a draft and published as a new version, and any previous version can be restored as a new draft.
• Webhook — A connection that lets Opal and an external system exchange events. An incoming webhook can trigger a task: Opal generates an endpoint, secured with a secret token, that another system calls to run it. An organization can also create outgoing webhooks that subscribe to platform events and send them to external systems.
• Widget — An interactive interface element an agent can present within a thread. A widget can display structured output and capture structured input through elements such as input fields and buttons, and interacting with it can invoke agent actions such as tool calls. Widgets are attached to agents and agent templates.
• Zero data retention (ZDR) — A data-handling arrangement in which the model provider does not retain the content sent to it. Opal enables zero data retention automatically whenever a model that supports it is used; it is available for select models and providers only, and the current policy details are published on Opal's website.