Impact at a Glance
- A small team shipped a hosted email identity for every user without managing MX records, SMTP infrastructure, or Google’s restricted scope review
- Every inbound email becomes a live agent turn
- Dana operates as two distinct identities — the user’s connected account and a dedicated mailbox for each Dana agent
About Dana Intelligence
Dana Intelligence is building a personal AI assistant that users access through WhatsApp, Telegram, or Slack. Users text Dana or send voice notes; Dana reads and replies to email in the user’s voice, manages their calendar, and books meetings and reservations without leaving the messaging app.
The product operates in two modes. Dana as you acts from the user’s connected Google or Microsoft account. Dana as Dana acts from a dedicated hosted address, provisioned at signup on the @mydana.ai domain. Both modes run over the same codebase. Both return a Nylas grant_id.
I wanted inbound email for Dana. A real address the agent could act from. What I want it to do is act like my executive assistant: manage a three-way meeting setup, send on my behalf, book the reservation. That’s the end state.
Brad Lindenberg
Founder @ Dana Intelligence
Dana Intelligence was founded by Brad Lindenberg, who previously co-founded Quadpay, the buy-now-pay-later platform acquired by Zip in 2020.
The Challenge: Email as a Live Channel
When a message arrives in Dana’s inbox, it becomes the next prompt in an active tool-calling session. The agent reads the email, decides what to do, and acts. That required treating inbound email as a live input channel, not a record to be polled.
That framing created two infrastructure problems simultaneously. The first was user-side: how to authenticate users into their own Google and Microsoft accounts cleanly enough that a small team could ship it. Google’s restricted scope review process takes months, requires a third-party security audit, and costs several thousand dollars annually. Building and maintaining that infrastructure independently was not a realistic option for an early-stage team.
The second problem was agent-side: how to give Dana its own inbox. Provisioning Google Workspace accounts for each user would have run into the same restricted scope review. Building inbound mail handling from scratch meant managing MX records, SMTP infrastructure, threading logic, and webhook delivery.
Dana Intelligence needed both problems solved on the same platform.
Why Nylas: Two Problems, One API
Nylas covers the user side and the agent side through a single integration. On the user side, Nylas handles OAuth into Google and Microsoft, normalizes threading and provider differences, and manages rate limits. On the agent side, Agent Accounts provisions a hosted mailbox on Dana Intelligence’s custom domain with a single API call, returning a grant_id that works against every Nylas endpoint.
Custom domains are registered in the Nylas dashboard. Nylas generates the MX and TXT records, which Dana Intelligence publishes at its DNS provider. When a new user signs up, Dana Intelligence calls the Agent Accounts endpoint and that user gets a dedicated address, a dedicated mailbox, and a dedicated calendar on @mydana.ai. The whole setup skips Google Workspace and the restricted scope review entirely.
Nylas describes this as a first-class identity: a dedicated mailbox and calendar that sends and receives like any human-operated account. Dana Intelligence runs the standard multi-tenant pattern Nylas documents for Agent Accounts: one Nylas application, one Agent Account per user, the same code path across the entire user base.
How It Works: Email as an Agent Turn
When a message arrives, Nylas fires a message.created webhook to Dana’s runtime. That event becomes the next turn in Dana’s OpenClaw tool-calling session. Dana reads the email body as a prompt, decides which tool to invoke, and acts.
If the user sends a voice note saying “Reply to Mike, tell him Thursday works, suggest Balthazar at 12:30,” Dana finds the thread, drafts a reply in the user’s voice, and queues it for approval. The user replies “send it” from Telegram. Dana sends from the user’s connected work account.
The reply arrives back through the same webhook pipeline. Dana picks up the inbound message as the next turn: surface it as a voice note, draft a follow-up, check calendar availability, or book the reservation. The email body is the prompt. Dana selects the tool based on intent.
Dana as Dana runs the same loop from a different identity. Both modes run through Nylas endpoints against their respective grant_ids. The specific capabilities in use:
- One Agent Account per user: Provisioned in a single API call, returns a grant_id
- Hosted mailbox on custom domain: MX and TXT records configured through the Nylas dashboard
- message.created webhooks: Inbound email delivered as a live trigger to Dana’s tool-calling session
- Drafts API: Dana drafts replies; the user approves in Telegram or Slack before sending
- Calendar API: Free/busy lookups, event creation, and conferencing links in the same identity
- OAuth via Connect: User-side authentication into Google and Microsoft without managing the restricted scope review
What’s Next
Dana for Work is the next release: a Slack-first enterprise version sold to teams, with Notetaker and Scheduler integrations on the roadmap.
The broader bet Dana Intelligence is making is that an AI assistant needs a real communications identity to function as a real assistant. Not a bot that reads your inbox when asked, but an agent with a dedicated address, a dedicated calendar, and the ability to act on both, autonomously, within boundaries the user sets. Agent Accounts is what makes that architecture possible without building the identity layer from scratch. That’s the foundation Dana is building on.