How Do I Set Up My First AI Workflow for My Small Business?

Summary

The steps

1. 1

   Connect an AI provider

   Before any agent can run in Dave by voolama (hellodave.ai), at least one AI provider must be configured. A provider is a named record that stores your API key, the service endpoint, and a list of available models. API keys are stored with AES-256-GCM authenticated encryption (NIST SP 800-38D, csrc.nist.gov/pubs/sp/800/38/d/final): the GCM authentication tag detects and rejects any tampering with the ciphertext. Keys are never returned through the API once saved. Sourced from the admin-providers help documentation at adminrodeo.qubiccontent.com/dashboard/help/admin-providers, DEV v0.11.0, confirmed 2026-08-30.

   Provider type	Default base URI	Where to get your API key	Model fetch?
   OpenAI	https://api.openai.com/v1	platform.openai.com, API keys section	Yes
   Anthropic	https://api.anthropic.com/v1	console.anthropic.com, API keys section	Yes
   Azure OpenAI	Your Azure endpoint (you supply this)	Azure portal, your OpenAI resource	Yes (requires a deployed model)
   Google AI	https://generativelanguage.googleapis.com/v1beta	aistudio.google.com	Yes
   Local (Ollama)	http://localhost:11434/v1	No key needed for local Ollama	Yes
   Custom	You supply the endpoint	Depends on the service	No (enter models as JSON)

   Only the Admin role can add providers. Dave by voolama has six default roles: Admin, Create, Curate, Use, Reporting, and Financial. Roles are editable.

   Do this

   • Navigate to Admin, then API Management, then Providers.
   • Click Add Provider. A form appears above the provider list.
   • Enter a descriptive name such as "OpenAI Production".
   • Select the Provider Type. The Base URI field auto-fills for known types.
   • Paste your API key. The field is masked by default.
   • Click Fetch Models. Dave by voolama calls the AI service and returns available models. Check the models you want, or click All.
   • Click Create Provider. The provider card appears in the list.

   Example

   A hypothetical marketing agency owner selects OpenAI as the provider type. The base URI auto-fills as https://api.openai.com/v1. She pastes her OpenAI API key, clicks Fetch Models, and sees gpt-4o and gpt-4o-mini. She checks both and clicks Create Provider. The provider card shows "OpenAI Production" with two models. See the Add an AI Provider quick-reference for a step-by-step walkthrough.

   Best practice

   The admin-providers help documentation recommends: 'Test connectivity after setup. After creating or editing a provider, use Fetch Models to verify that the API key and base URI are correct.' If Fetch Models returns an error, the most common cause is an incorrect API key. Re-enter the key and try again. Name providers by purpose and environment: "OpenAI Production" rather than "My Provider".

2. 2

   Create an agent with a system prompt

   An agent in Dave by voolama (hellodave.ai) is a reusable AI component that wraps a provider with a system prompt, a model choice, and behavior settings. Every Agent Interaction node in a workflow references an agent. You create the agent once and reuse it across multiple workflows. Sourced from the getting-started help documentation, DEV v0.11.0, confirmed 2026-08-30.

   Only the Admin and Create roles can create agents. The Curate role can view agents but cannot create or modify them.

   Do this

   • Navigate to Agents in the sidebar. Click Create Agent.
   • Give the agent a name that describes its function: "Social Caption Drafter" rather than "GPT-4o Agent". Model names change; functions do not.
   • Add an optional description and tags, then click Save.
   • Click into the agent to open its detail page. Click Create Version.
   • In the Config JSON editor, enter a JSON object. Example: {"systemPrompt": "You are a social media copywriter for a small business. Write one Instagram caption of 50 words or fewer based on the brief provided. Use a friendly, direct tone. Do not use hashtags.", "model": "gpt-4o", "temperature": 0.7, "maxTokens": 200}
   • Validate the JSON in a browser console using JSON.parse() before submitting. A SyntaxError means a formatting problem to fix first.
   • Click Create Version. The version appears as v1.

   Example

   The hypothetical agency owner creates an agent named "Social Caption Drafter". She enters a system prompt instructing the AI to write one Instagram caption of 50 words or fewer, friendly tone, no hashtags. She sets model to gpt-4o and temperature to 0.7. She validates the JSON in her browser console, then clicks Create Version. The agent is now at v1.

   Best practice

   Write the system prompt as a specific, testable instruction. "Write one Instagram caption of 50 words or fewer" is testable. "Write good social media content" is not. A specific prompt produces consistent output that is easier to review and approve under EU AI Act Article 14(4) human oversight requirements.

3. 3

   Build a workflow with a Human Review step

   A workflow in Dave by voolama (hellodave.ai) is a visual graph of nodes connected by edges. The Workflow Editor is a drag-and-drop canvas: no code required. Sourced from the workflow-editor help documentation at adminrodeo.qubiccontent.com/dashboard/help/workflow-editor, DEV v0.11.0, confirmed 2026-08-30.

   Only the Admin and Curate roles can open the Workflow Editor and save workflow versions.

   Node type	What it does	Creates a task?	Requires an agent?
   Start	Entry point. Every workflow must have exactly one.	No	No
   User Interaction	Pauses workflow, creates a task for a person to submit input.	Yes	No
   Agent Interaction	Invokes an AI agent automatically. No human involvement.	No	Yes
   Human Review	Pauses workflow, creates a task for a person to approve, reject, or request changes. Satisfies EU AI Act Article 14(4) human oversight requirement.	Yes	No
   End	Terminal node. A workflow can have multiple End nodes for different outcomes.	No	No

   Do this

   • Navigate to Workflows in the sidebar. Click Create Workflow. Name it "Social Caption Pipeline" and click Save.
   • Click into the workflow to open the Workflow Editor.
   • From the Node Palette on the left, drag a Start node onto the canvas.
   • Drag a User Interaction node below Start. Rename it "Collect Brief" in the Properties Panel on the right.
   • Drag an Agent Interaction node below. Rename it "Draft Caption". In the Properties Panel, select the "Social Caption Drafter" agent and its v1 version.
   • Drag a Human Review node below. Rename it "Owner Review".
   • Drag an End node below.
   • Connect the nodes: hover over the bottom handle (source) of Start, drag to the top handle (target) of Collect Brief. Repeat for each pair in sequence.
   • Click Save Version. The workflow is now at v1.

   Example

   The hypothetical agency owner drags five nodes onto the canvas: Start, Collect Brief (User Interaction), Draft Caption (Agent Interaction, Social Caption Drafter v1), Owner Review (Human Review), and End. She connects them in sequence and clicks Save Version. The node/edge counter shows "5 nodes, 4 edges". See the Build Your First Workflow quick-reference for a visual walkthrough.

   Best practice

   The workflow-editor help documentation states: 'Always start with a Start node. Every workflow requires exactly one Start node as its entry point.' It also states: 'Ensure all paths lead to an End node. Dangling paths may cause instances to hang at runtime.' Save frequently: each save creates a new immutable version, so there is no risk of overwriting your work.

4. 4

   Launch an instance and approve the output

   An instance is a single run of a specific workflow version in Dave by voolama (hellodave.ai). At each human node (User Interaction or Human Review), the instance pauses and creates a task in the task inbox. Completing the task resumes the instance. Sourced from the getting-started help documentation, DEV v0.11.0, confirmed 2026-08-30.

   Only the Admin and Use roles can launch instances and complete tasks.

   Do this

   • Navigate to Instances in the sidebar. Click Create Instance.
   • Select "Social Caption Pipeline" and its v1 version. Click Launch.
   • Navigate to Tasks. Open the Collect Brief task. Type a brief and click Submit.
   • The instance calls the Social Caption Drafter agent automatically. Refresh Tasks after a few seconds.
   • Open the Owner Review task. Read the AI-drafted caption in the task payload.
   • Click Approve if the caption is good. Click Request Changes and add a note if it needs revision. Click Reject if it is wrong.
   • Navigate to Instances. The instance status shows completed (if you approved).

   Example

   The hypothetical agency owner launches Social Caption Pipeline v1. She opens the Collect Brief task and types: "We just launched a new summer menu. Highlight the lemon tart." Thirty seconds later an Owner Review task appears. The AI has drafted: "Summer is here and so is our lemon tart. Light, bright, and made fresh every morning. Come in and try it this week." She clicks Approve. The instance status changes to completed. See the Launch an Instance and Work a Task from Your Inbox quick-references.

   Best practice

   The getting-started help documentation recommends: 'Test with a simple workflow first. Before building a complex pipeline, create a minimal Start, Agent Interaction, End workflow and run an instance to verify your agent config works.' If the AI output is not what you expected, the most common fix is to make the system prompt more specific. Return to the agent, create a new version with an updated system prompt, and run another instance. The previous version is preserved.

What you need before you start

Roles and who does what in Dave by voolama for a small team

What this guide does not cover