Dave is a blank canvas with real guardrails. Here's what teams put on it first.
Join the WaitlistResearch → draft → review → publish, as one auditable pipeline. AI agents write on the model you pick, human editors approve in a review flyout, and API-call nodes push the finished piece straight into your CMS. Every version tracked; nothing goes live without sign-off. (This site's content runs exactly this way.)
Trade email threads and 'who approved this?' for structured, logged approval flows: role-based review nodes, voting, full reviewer context, and a complete audit trail, with SOC 2 (7-year retention) and GDPR compliance modes for work that's regulated or brand-critical.
Point agents at a question; they gather, summarize, and structure the answer, then hand it to a human to validate before anyone trusts it. Export to JSON, HTML, or plain text and feed it downstream. A research backlog becomes a button.
One deployment, many clients or business units: full tenant isolation, per-tenant branding and encryption, feature flags, white-label from day one. It's literally how voolama runs five brands on a single Dave instance.
Wire Dave to any REST API (CRM, tickets, billing, webhooks) across 129 endpoints, and gate the steps that move money or carry risk. Automate the repetitive; review the consequential.
A real example: an AI-drafted, human-approved content pipeline. The same pattern teams reuse for research, approvals, and back-office tasks.
The instance launches: scheduled, triggered via the API, or kicked off manually.
An Agent Interaction node drafts the content, with up to 10 retries configured before it ever reaches a human.
A Human Review node routes to the assigned reviewer. Approve moves it forward; Reject requires a written reason and sends it back.
Approved output flows to the next step: an API Call node pushing it into your CMS, ticketing system, or wherever it needs to land.
| Role | What they can do |
|---|---|
| Admin | Full tenant administration: every permission, unrestricted. |
| Create | Build and manage AI agents: prompts, models, configuration. |
| Curate | Design workflows in the visual editor, the graph itself. |
| Use | Run instances and complete tasks: open assigned work, provide input, approve or reject. |
| Reporting | Analytics only: run volume, failure rates, latency, and SLA metrics. No edit access. |
