For your first CampaignA developer's data-production project and its settings, follow Start with a data need. Return here when choosing a different authoring route.
Choose your starting point
| Your goal | Start here |
|---|---|
| Collect original responses, recordings, observations, or evidence from people | Create a human-sourcing Campaign below. |
| Add human verification or judgment to an automated process | See the agentic data-sourcing example below. |
| Label or validate existing records | Configure a Supply step with the appropriate task type and supported inputs. |
| Retrieve data that already exists | Follow Get existing data. |
| Work on a Campaign your team already owns | Join its organization, open the Campaign, and confirm your Campaign role. |
Explore the Developer portal
Builder brings the workflow canvas and Tiny-h, the portal's assistant, into the same workspace. Describe the data you need, ask about the steps on your canvas, or use DSA's deep analysis to examine a data question. Where dataset analysis is available, you can attach an existing dataset for Tiny-h to analyse.
Tiny-h can propose changes to the Campaign. Read the proposed changes and choose Apply to accept them. A response in the conversation does not, by itself, change the workflow or launch the Campaign.
Use the preview to inspect four parts of authoring:
| Part | What you decide |
|---|---|
| Workflow | What people and automated steps do, how results are checked, and where accepted data goes. |
| Entry Gate | Who can join the Campaign and which participation requirements they must meet. |
| Agreements | Which published project agreements Contributors must accept. The platform's required agreements are shown as locked; the applicable agreement depends on the Contributor's actor type. |
| Qualifications | Which profile requirements the work needs, such as language proficiency or location. Leave unrelated requirements unset. If training is needed, choose an Academy course, its published version, and the required completion state. |
Participation requirements and result checks serve different purposes: the Entry GateThe workflow step that holds a Campaign's joining requirements controls who may join; the workflow checks the work they submit. Requiring an agreement or qualification does not make every submissionThe record created when work is submitted acceptable.
What can you build with a Campaign?
Human-sourced data
Ask people to create or collect data that your application needs: original sentences, speech samples, local observations, or source-backed research.
For example, collect three original sentences and ask each Contributor to classify their sentence as a question or statement. Review the answers and retrieve the accepted records for your application.
Example path: Entry Gate → Supply (source) → Accept → Sink.
Agentic data sourcing with human verification
Combine automated checks with work that benefits from a person's judgment. For example, Contributors find source material; a configured Agent checks it against the research brief; another human step verifies the claims and adds evidence; accepted results are mapped into the fields your application needs.
Example path: Entry Gate → Supply (source) → Agent TaskA specific item a contributor claims or enters → Supply (validate) → Accept → ETL → Sink.
This is a workflow recipe, not a built-in preset. An Agent Task needs an Agent configured in the organizationThe Developer-side unit that holds Campaigns, data access, and funds, instructions, and downstream routing. Configure the validation step against its source work, and define what happens when checks fail. Do not treat an automated check as proof that a claim is true.
If an external agent discovers candidate records, confirm the supported input route before designing around it. The documented dataset-import route accepts eligible datasets managed by the same organization; it is not a general web-crawling or arbitrary-JSON ingestion endpoint.
Human labeling and evaluation
Ask Contributors to categorize records, compare responses, or explain why an answer meets a rubric. A human step can capture a label and written reasoning, followed by a defined review and acceptance process.
Use a Supply step with a supported task type and form. The standalone Evaluate node is currently locked; a human evaluation task does not require that node.
For more patterns and setup considerations, see Campaign use cases.
Choose how to author the Campaign
You can prepare the same Campaign through different authoring routes:
| Route | Best when | How to begin |
|---|---|---|
| Builder with Tiny-h / DSA | You want help developing a brief or reviewing the design | Describe your data need in Tiny-h, inspect its analysis and any proposed changes, then apply the changes you choose. Continue configuring and testing the draft in Builder. |
| Builder: start from a preset | You want a guided starting structure | Open Campaigns → Open Builder, choose Start from a preset, then adapt the form, workflow, rewards, and requirements. |
| Builder: start from a blank canvas | You want to design the workflow visually | Choose Start from a blank canvas and add the Entry Gate first. |
| CLI | You want a repeatable workflow with a Campaign File | Start from a template or existing version, then validate, plan, and apply the candidate with the Humanbased CLI. |
| Campaign API | Your application prepares or updates Campaigns | Use the Campaign API to validate, plan, and apply a candidate under the organization's grants. |
| Coding agent via local MCP | You want an agent client to assist with supported CLI operations | Use a CLI build that includes hbd mcp serve; see CLI agent integration. |
First Campaign in a new organization? Start in the portal. Campaign-authoring and launch grants on an API key require its creator to be an editor or owner of at least one Campaign. The CLI and API routes therefore have an initial setup prerequisite.
Creating or saving a draft does not launch it. API launches also require a person's confirmation in the portal. An agent's access to delivered dataThe authorized output a Developer can retrieve, with its own delivery and access state is separate from permission to author Campaigns.
The remaining steps use Builder so you can see the complete first-result path.