Domo

Each agent gets a Tools tab where the builder picks Domo global functions, external global functions, custom Code Engine functions written in JavaScript or Python, or whole workflows already built in the instance. Tool and parameter descriptions tell the model when to call each one, and a static value can pin any input.

Code Engine functions reach outside systems through stored account credentials, so an agent can update a CRM record, open a ticket or send a notification, and more than 1,000 connectors including Salesforce, SAP and Snowflake bring data in. A lead qualification example assigns a rep and emails the details in one run.

Sourcedomo.com/docs/s/article/000005865read 2026-10-18

Domo Workflows models a business process as shapes and flows on a canvas, with decision logic, user tasks, timers, subflows, versioning and a Start Workflow step that chains one workflow into another. An agent task is one shape in that process: it reasons over its inputs, calls tools, and writes typed outputs that later shapes map into the next step. Executions can be retried or recovered from a failed shape, and a workflow can itself be handed to an agent as a tool, so a single agent can launch a whole multistep process. Each agent request has a one hour limit.

Sourcedomo.com/docs/s/article/000005108read 2026-10-18

Agents ground on the customer's governed DataSets, which Domo keeps refreshed from its connectors, and on FileSets, document collections that agents and assistants query in natural language. The Knowledge tab of an agent takes up to five DataSets, five FileSets and five files, and agent knowledge is indexed documents served through retrieval augmented generation. A text embedding service in the AI Service Layer powers semantic search, and text to SQL lets an agent query structured data directly. Personalized data permissions continue to apply to what each user's agent can read.

Sourcedomo.com/docs/s/article/agent-vs-tool-knowledgeread 2026-10-18

User tasks put a person into any workflow, either through a form assigned to a queue in Task Center, including its mobile app, or through an email with response options embedded in the message. The workflow waits for that answer before the next step runs, and interrupting or recurring timers on a task escalate or reroute work that sits too long.

Agent outputs can feed a review task before anything is written or sent, and the catalogue's fraud and anomaly agents route flagged items through approval steps before tickets or holds are created. Tool inputs can be pinned to static values so the agent cannot change them.

Sourcedomo.com/docs/s/article/000005797read 2026-10-18

Single sign on runs over SAML 2.0 or OpenID Connect, with Okta, Entra ID and Google guides, and SCIM provisions users and groups. Custom roles are built from grants, with separate grants to use AI services, use AI Chat, manage conversational agents and manage toolkits, and personalized data permissions restrict rows per user. Customers can supply and rotate their own encryption keys, and MCP tools act only within the connected account's permissions. Current certifications are SOC 1, SOC 2, ISO 27001, ISO 27018, HIPAA and HITRUST, alongside GDPR and CCPA programs and annual penetration tests.

Sourcedomo.com/platform/securityread 2026-10-18

Every workflow run, including its agent steps, appears in an Executions list with status, and opening one replays it on the canvas, showing how long each shape took, where a run failed and the data passed through each step in a panel below. Failed runs can be recovered or retried with edited execution data. Separately, the instance Activity Log records user and system events, filters by time and actor, exports to CSV or a webhook, and feeds a DomoStats dataset for long term reporting. Execution access is governed by its own permission on each workflow.

Sourcedomo.com/docs/s/article/000005179read 2026-10-18

An agent task receives its prompt, instructions and variables when a workflow execution reaches it and returns its outputs to that execution, and nothing in the agent setup holds what one run learned for the next. Standing guidance lives in the agent's role and instructions, which people write.

Longer lived facts sit in DataSets, AppDB collections or FileSets, the customer's own business data, which an agent reads like any other source and can update through a function. Workflow variables carry state from step to step inside a single run, and the execution record keeps those values for review afterward.

Sourcedomo.com/docs/s/article/000005865read 2026-10-18

Domo is a hosted service. Its instances run on Amazon Web Services in the eastern United States, Australia, Ireland, Canada, Japan and India, with Microsoft Azure hosting in the eastern United States, which places the platform in several geographies. A buyer gets no stated way to pick one of those regions for its instance. There is no self hosted or customer tenant edition, and the custom plan's private link connectivity and HIPAA compliant environment both sit inside Domo's own hosting. Domo can be bought through the major cloud marketplaces against existing cloud commitments.

Sourcedomo.com/company/subprocessorsread 2026-10-18

A library of example agents spans departments, categories and integrations, among them retail promotion effectiveness, manufacturing transformation, fraud and transaction risk, waste pattern detection, lease extraction and P&L analysis, each offered as an expert built starting point for custom work. Workflows can start from a template library, and the AI Library, in beta, holds Toolkits Domo provides for common use cases alongside customer built ones. AgentGuide recommends a starting point from a short questionnaire. Each of these is a base the customer configures on its own data.

Sourcedomo.com/agentcatalystread 2026-10-18

A deployed workflow starts from a timer that runs every minute, every five or fifteen minutes, hourly, daily, on weekdays, weekly or monthly, from a DataSet alert when a metric crosses its threshold, from a widget in an App Studio app or a pro code app, through an API trigger with generated JavaScript and Python calls, when someone requests access to a card, page or app, or from another workflow. An MCP trigger is switched on automatically so an outside AI assistant can start a workflow, and manual runs work from the canvas.

Sourcedomo.com/docs/s/article/000005179read 2026-10-18

Admins set the default model for each of twelve AI services, from chat completion and tool calling to text to SQL, planning and embeddings, in the AI Service Layer settings. Bring Your Own Model imports models through a connected account, with Meta Llama on Amazon Bedrock as the worked example, and an imported model can become the default for a service or stay available for programmatic use. DomoGPT, Domo's hosted model, is the out of the box default, and an individual agent task can switch to any model the admin has made available.

Sourcedomo.com/docs/s/article/bring-your-own-modelread 2026-10-18

The developer portal carries OpenAPI specifications for DataSets, streams, users, groups, activity log, AI services, Code Engine, Task Center, workflows and roles and governance, with an API trigger that starts a specific workflow version. The Domo MCP server, reachable with standard OAuth from Claude, ChatGPT, Gemini Enterprise or Amazon Quick Suite, answers questions across the platform and builds cards, App Studio apps, workflows, workspaces and ETL pipelines from a conversation, with toolkits for alerts, Code Engine, Task Center queues and user administration.

Sourcedomo.com/docs/s/article/domo-mcpread 2026-10-18

The agent builder has a Test Agent button that runs the agent with sample values for each prompt variable and shows its output before the step is saved. Workflows Test Run executes a version that has not been deployed, highlights each shape green, blue or red as it completes, runs or fails, and keeps the run under a Test Run tag in the executions list for later review. Both are one off checks a builder reads by eye. There is no set of saved test cases with expected answers or a score to compare one agent version against another.

Sourcedomo.com/docs/s/article/000005842read 2026-10-18

Agents act through functions, APIs, connectors and workflows, the tools an agent invokes against outside systems: Domo global functions, external functions, Code Engine functions in JavaScript or Python, and whole workflows.

No hosted browser, remote desktop or screen control is part of the agent toolset, and the catalogue agents work on DataSets, FileSets and connector data. Work in an application without an API would need a custom Code Engine function that calls whatever endpoint exists, or a separate automation product. App Studio and pro code apps are where people see and act on agent output, and Domo users build those apps by hand.

Sourcedomo.com/docs/s/article/tools-vs-skillsread 2026-10-18