Glossary
A reference for common Qore terms.
Reference entries for UI terms, acronyms, and platform-specific vocabulary used throughout the builder, plus richer concept entries for the domain ideas the screens reference. Reference entries are short (one to three sentences) and point to the page where the term is most fully discussed. Concept entries are longer (two to four sentences) and carry the teaching weight the screen pages keep light.
Entries are alphabetized. The page is intended for lookup — search the page, or follow an in-page anchor link from another page in the guide.
A
Agent Builder
The UPTIQ surface you use to configure one AI agent — its identity, instructions, attached resources, runtime settings, and integrations. Reached at /agent-builder/<agentId>. See Get Started.
AI Gateway
The platform layer that every model request passes through. It enforces organization-level policies, including approved LLM families and providers.
Agent Capability
A dropdown on Agent Behavior that controls how deeply the agent thinks about each request. Default is Auto. Higher settings produce more thorough reasoning at higher latency and cost.
Agent Output Format
A dropdown on Basic Details that controls how the agent's text responses are rendered. Options are Markdown and Json. Use Markdown for responses read by humans. Use Json when the host application parses the response.
Agent Preview
The chat-style left pane in the builder, where you can talk to the current draft of the agent without leaving the page. See Get Started.
API Call (skill)
A first-party skill type in the Skill Library that lets the agent reach external HTTP APIs at runtime. Use it for services not covered by Apps & Services connectors. Reached from Adding a skill to the agent.
API Integration
One of the three deployment options for integrating an agent. See the REST API tab on Agent Integration. Produces credentials and endpoints for programmatic, server-side invocation of the agent.
App (in UPTIQ)
The container that an agent lives inside. Each agent is scoped to one app; the builder URL carries the appId query parameter. Apps are themselves scoped to an account.
App workspace
The editor for one app. It includes the build conversation, live preview, and action bar. See App Workspace.
Autonomy
A slider on Agent Behavior controlling how independently the agent commits to multi-step plans. Marked Low / Med / High; default Med. Higher autonomy means fewer confirmations between steps.
B
Basic Details
The first section in Manage Agent, holding the agent's identity — avatar, name, description, and output format.
Build version
Each app build is a numbered version tied to a commit. The version selector marks the newest build as Latest and the live build as Published. Testing, security, and data flow reports each apply to one version.
C
Checklist
A pre-built execution plan — a named sequence of skill calls the agent runs end-to-end when a request matches the checklist's criteria. See Execution Plan/Checklist for the management page. See the Planner entry below for how a checklist fits into runtime execution.
CoT Reasoning
A toggle on Agent Behavior. When on, the agent generates a visible chain of thought as part of its response. See the Chain of Thought entry below.
Chain of Thought
A reasoning approach where a language model produces, in addition to its final answer, the intermediate steps it took to reach the answer. Surfacing the chain of thought makes the agent's reasoning auditable — useful for high-stakes or compliance-sensitive tasks where the how matters as much as the what. The platform exposes this as the CoT Reasoning toggle. The disabled COT nav entry is likely a richer view of the same concept that hasn't yet shipped.
Connect (button)
The action revealed on hover over an integration row in Apps & Services. Initiates the integration's connection flow, typically OAuth, so the agent can call the service at runtime.
Connect BYOK
One of the three connect methods in Connect Methods. Short for Bring Your Own Key. You supply the integration's credentials directly — usually an API key, OAuth client, or webhook secret — and the platform stores them encrypted for runtime use.
Connect Method
The authentication and connection pattern an integration uses in the Integrations Hub. Every integration uses one of three connect methods — Connect BYOK, Connect via UDG, or MCP Sandbox Account. See Connect Methods.
Connect via UDG
One of the three connect methods in Connect Methods. The integration authenticates through UDG instead of directly with the provider. Use it for partner-managed integrations where Uptiq handles the connection layer.
Custom Trigger
The detail form for editing or creating one trigger, reached from the expanded Agent Triggers modal. Carries three required fields: Trigger Source, Trigger Name, and Description. See Triggers.
Control Center
The Qore surface for monitoring and governing running apps account-wide. Studio is where you build apps and agents.
CVE / GHSA
Public advisory identifiers for vulnerabilities. CVE means Common Vulnerabilities and Exposures. GHSA means GitHub Security Advisory.
D
DLP
Short for Data Loss Prevention. Google Cloud DLP detects sensitive data from a large catalog, including custom infoTypes.
Dataset
A collection of test cases used to evaluate one agent. In Evals, datasets can be created manually or generated with AI, and each agent can have up to 100 of them.
Draft Version
The header-strip popover that saves a snapshot of the current agent configuration as a recoverable version. See The Builder Canvas. Distinct from Executor Version, which is the platform runtime version the agent is pinned to.
E
Embed Agent
One of the three deployment options for integrating an agent. See the Embed widget tab on Agent Integration. Produces an embeddable widget that surfaces the agent as a chat on a web page.
Enforced (guardrail mode)
A guardrail mode under which the guardrail blocks execution when its check fails. Contrast with Monitoring mode, which logs but does not block. See Guardrails (Beta).
Eval Run
One execution of a dataset against a live agent in Evals. An eval run tracks status, progress, per-test-case results, scores, and any generated suggestions.
Evals
The platform's offline evaluation framework for testing agent quality against scored datasets before deployment. Use it to build datasets, run them against a live agent, review scores, and apply AI-generated improvements. See Evals.
Executor Version
The platform runtime version the agent is pinned to. Set on Runtime & Versioning. Controls which features are available — for example, PII Tokenization is gated by executor version.
Export Agent
The action on Export Agent that packages the agent's full configuration into a portable bundle for reuse or deployment elsewhere.
F
FAQ memory
A toggle on Agent Behavior controlling whether the agent uses stored FAQs and their canonical answers when a question matches a known FAQ. Default on.
G
Guardrail
A validator that inspects content as it flows through the agent's execution pipeline. The platform organizes guardrails into four execution layers — input validation, pre-skill execution, post-skill execution, and output validation — and routes content through every active guardrail at the appropriate layer. A guardrail can pass content through, modify it, or block it. See Guardrails (Beta).
H
Headless Agent Integration
One of the three deployment options for integrating an agent. See the JS SDK tab on Agent Integration. The lowest-level invocation pattern — server-to-server, without UI scaffolding.
I
Improve with AI
A button above the Instructions editor. Runs an AI pass over the current instructions and proposes improvements for clarity, structure, and consistency.
Integrations Hub
The centralized catalog and management surface for third-party integrations in Uptiq. Use it to discover available integrations, configure credentials, and make integrations available to agents, skills, workflows, and apps. See Integrations Hub.
infoType
Google Cloud DLP’s name for a detectable sensitive-data type. It includes built-in types, such as card numbers and emails, plus custom types.
J
JavaScript Code (skill)
A first-party skill type in the Skill Library that lets the agent execute JavaScript code at runtime. Useful for custom logic that doesn't fit other skill types. Reached from Adding a skill to the agent.
K
Knowledge
The category of content the agent can read from at runtime — documents, files, and reference material — managed through the Add Knowledge action on the canvas. See Attaching Knowledge to an Agent.
L
Layer 1: Input Validation
The first execution layer in Guardrails (Beta). Runs on the inbound user message before any other processing.
Layer 2: Pre-Skill Execution
The second execution layer in Guardrails (Beta). Runs before each skill is invoked by the planner.
Layer 3: Post-Skill Execution
The third execution layer in Guardrails (Beta). Runs after each skill completes, on the skill's output.
Layer 4: Output Validation
The fourth execution layer in Guardrails (Beta). Runs on the final agent response before it leaves the agent.
Layout (tab)
The default tab on The Builder Canvas. Shows the agent and its attached resources as a node graph.
LLM
Short for large language model. An LLM reasons over requests and generates agent responses. Organization policies control which LLM families and providers an agent may use.
M
Manage Agent
The configuration dialog that holds every substantive setting for the agent — basic details, instructions, behavior, skills, triggers, security, monitoring. Reached from the Manage Agent button in the header strip.
Markdown (output format)
One of two options for the Agent Output Format. The agent's text responses render as Markdown — headings, bullets, code blocks, and links all formatted.
Masking provider
The tenant-level service that detects and tokenizes personally identifying information before it reaches the agent's model. The provider determines what entity types can be detected, which in turn determines what entity types can appear on the PII Tokenization exclusion list. If a tenant has no masking provider, the page shows the provider as Off.
Model family
A group of related models, such as GPT or Claude. An organization can restrict an agent to approved model families.
Model provider
The vendor that hosts a model, such as OpenAI, Anthropic, AWS Bedrock, or Azure OpenAI. An organization can restrict an agent to approved providers.
MCP
Short for Model Context Protocol. An open protocol for connecting language models to external tools and data sources through a standardized interface. The platform exposes MCP servers as a catalog under MCP Tools; connecting an MCP server adds its tool set to the agent's reach. The same protocol underlies several Skill Library entries and the broader marketplace tabs in Adding a skill to the agent.
MCP Marketplace
A filter tab in Adding a skill to the agent that lists third-party MCP servers beyond the first-party Skill Library and the Uptiq-curated marketplace.
MCP Sandbox Account
One of the three connect methods in Connect Methods. Used for banking-core and certain financial-platform integrations delivered through Kinective. The connection exposes provider operations as MCP tools the agent can call at runtime.
MCP Server (skill)
A first-party skill type in the Skill Library that lets the agent connect to a third-party MCP server. Reached from Adding a skill to the agent. Distinct from connecting an MCP through MCP Tools, which uses the catalog flow.
Memory
A toggle on Agent Behavior controlling whether the agent remembers user preferences across conversations. Default off. Distinct from FAQ memory.
Monitoring (guardrail mode)
A guardrail mode under which the guardrail runs its check but does not block execution on failure. Use as a staging step before promoting a new guardrail to Enforced mode. See Guardrails (Beta).
Monitoring (section)
The Manage Agent section that lists the agent's historical executions, one row per run. Distinct from the guardrail mode above. See Monitoring.
Managed DB
The Uptiq Managed DB is a PostgreSQL database that Qore runs for an app.
O
OnBuild
The trigger that runs automated tests when an app builds.
Online Evals
The platform's production-side evaluation system that scores real conversations in the background after an agent is deployed. It complements Evals, which validates quality offline before deployment. See Online-Eval.
P
Pre-attached Skills
The set of skills the platform attaches to every agent by default — display_to_user, reason, web_search, query_knowledge, list_files, and others. Visible on the Layout canvas as a card and in Skills & Capabilities as the top of the skill list.
PII tokenization
Replacing detected personally identifying values — names, account numbers, addresses, and emails — with opaque tokens before content reaches the agent's model, so the model never sees the underlying values. The platform performs this automatically through the tenant's Masking provider; the PII Tokenization page is the exclusion list on top of the automatic default. Tokenization is reversed downstream where authorized, so the user sees the original values in responses even if the model didn't.
PII
Short for personally identifying information. PII includes data that identifies a person, such as a Social Security number, card number, email address, or phone number.
Plan (tab)
The non-default tab on The Builder Canvas. Shows the agent's pre-built execution plan when Execution Plan/Checklist entries have been configured. Reads Plan is not available when none have.
Planner
The part of the agent runtime that decides what skills to call, in what order, for each incoming request. The planner reads the agent's instructions and the user's input, then picks one or more skills from Skills & Capabilities. When a Checklist matches the request, the planner follows the checklist's pre-built sequence; otherwise it generates a plan dynamically. Pair this with the Executor entry below.
Executor
The part of the agent runtime that actually invokes the skills the planner chose, passes their results back through the model, and renders the final response. The executor is what the Execution Model setting selects.
Planning Model
A dropdown on Agent Behavior selecting the model used for planning tasks. Can differ from the Execution Model; a common pattern is a heavier model for planning and a lighter model for execution.
Preprocessing Pipeline
A checkbox in the Attaching Knowledge to an Agent flow that runs additional preparation on uploaded content — chunking, OCR for scanned PDFs, and layout-aware extraction — before indexing. Leave off for clean text documents. Turn it on if retrieval quality is poor or the source is scanned.
Prompt Sanitization
A toggle on Agent Behavior that, when enabled, detects malicious user prompts — prompt injection and jailbreaks — before they reach the model. Currently shown as Not Supported in the example agent's configuration. The gate is not documented in the UI.
Platform package
An app exported from Qore. Importing a Platform package restores it as exported. Git repositories and projects from other builders are converted into Platform apps.
Preview build vs. published build
An app has a Preview build and a Published build. Preview is the build you edit in the workspace. Published is the build users see at the live URL. DB Explorer and environment variables can use either scope.
Python Code (skill)
A first-party skill type in the Skill Library that lets the agent execute Python code at runtime. Useful for custom logic that fits Python better than JavaScript. Reached from Adding a skill to the agent.
Q
query_knowledge
A pre-attached skill the agent uses to look up information from attached knowledge sources. Without query_knowledge attached, the agent can't reach attached knowledge — even if the knowledge has been imported. See Attaching Knowledge to an Agent.
R
reason
A pre-attached skill the agent uses for reasoning, decision-making, or summarization. See Skills & Capabilities.
Response Mode
A dropdown on Agent Behavior controlling the speed-versus-quality trade-off in responses. Default Advanced produces higher-quality but slower responses. Faster modes are available.
S
Secret
An encrypted value — API key, token, or password — stored on Secrets & Variables and referenced elsewhere as $secret.{name}. See also Variable.
Skill
A callable capability the agent can invoke at runtime — code, an API, a tool exposed by an MCP server, another agent, or anything cataloged in the skill library. Every skill the agent has appears in Skills & Capabilities.
Skill Library
A filter tab in Adding a skill to the agent that lists first-party skills authored by Uptiq — API Call, JavaScript Code, Python Code, MCP Server, Structured Data Query, and others.
SSN
Short for Social Security number. SSN is a US identity entity type available through the Regex masking provider.
Structured Data Query (skill)
A first-party skill type in the Skill Library that lets the agent ask questions of a database in natural language. The platform translates those questions to SQL and runs them. Reached from Adding a skill to the agent.
Suggested Changes
AI-generated recommendations produced by Evals after a run completes with failures or a low average score. Suggestions can target instructions, skills, knowledge, or guardrails, and can be applied back to the agent from the run results.
Studio
The Qore surface where you build apps and agents. App Builder is part of Studio.
T
Test Case
One evaluation item inside a dataset. In Evals, a test case includes a question, an expected answer, and optional document context or attachments.
Toggle Panel
The rectangular split icon in the header strip that hides or shows the Agent Preview pane. See The Builder Canvas.
Trigger
A named condition described in prose that, when met, fires the agent. Each trigger has a Source, Name, and Description. The description is interpreted by a runtime model rather than parsed as a rules expression. See Triggers.
Trigger Source
A required dropdown on the Custom Trigger form selecting the kind of event the trigger watches for. App is the only source observed in the example agent.
Trivy
The scanner that Security uses to find dependency vulnerabilities.
Try Agent
The green button in the header strip that initiates agent import into the Agent Preview pane. After import, the chat composer becomes active.
U
UDG
Short for Uptiq Data Gateway. A managed integration layer that brokers connections to partner services in the Integrations Hub. UDG handles connection infrastructure and provider compatibility for integrations that use Connect via UDG.
Uptiq MCP Marketplace
A filter tab in Adding a skill to the agent that lists MCP-server-backed skills curated by Uptiq. Distinct from the broader third-party MCP Marketplace tab.
V
Variable
A non-sensitive value stored on Secrets & Variables — base URLs, identifiers, and defaults — referenced elsewhere as $agent.{name}. See also Secret.
Vault
The secure store that maps an original PII value to its opaque token. Authorized downstream steps can restore the original value when needed.
W
web_search
A pre-attached skill the agent uses to retrieve current information from the web. Useful when the agent's training data may be out of date. See Skills & Capabilities.
Whitelisted Domains
The allowlist on Whitelisted Domains of web origins where the agent's embedded widget is permitted to run. Calls from origins not on the list are refused.
Widget Key
A key type in Authentication used by the agent's embedded widget. The widget includes the key in its calls to the platform. The key is scoped to allow the widget to run on the Whitelisted Domains.
X
x-platform-key
The request header that carries an API key for the App Builder MCP endpoint.
Last updated