AI agents & chatbot
In the previous lesson you added Storage in DYPAI in a simple way. This video is about building an AI agent for your app.
How agents work in DYPAI
There's no standalone "AI Agents" page. An agent is simply an endpoint that uses the agent workflow node, and your other endpoints can be marked as tools (is_tool=true) so the agent can call them. You build and configure all of this in the API workspace (Build → API).
DYPAI Managed vs BYOK
Before you build, decide how the agent gets its model:
- DYPAI Managed (the zero-config default) — no API key needed. Usage is metered as AI Credits, and the models you can pick from are plan-gated, managed at the org AI Models page (
/{org}/ai-models). - BYOK (bring your own key) — store your own OpenAI, Anthropic, or Gemini key under Credentials, and you pay the provider directly.
For your first agent, the simplest path is DYPAI Managed: there's nothing to set up. If you'd rather use your own Gemini key (as the video does), add it under Credentials → Add new first, then point the agent at it.
The assistant endpoint
Open a new conversation and ask the AI to implement an AI assistant for your site (a chat you can talk to). The agent will check for any credentials you added, scan nodes and documentation, and start generating the right endpoint — one built around the agent node — for the assistant.
First tests
When the run finishes, try a simple prompt—for example “hello” or “how do I create a task?” You should get an answer because the stack is already wired to DYPAI Web. By default the assistant may pick a baseline model; you can change it.
Changing the model
Open the agent endpoint in the API workspace and edit the agent node's model. With DYPAI Managed, the available models come from your org AI Models page (plan-gated). For example you can switch to a newer Gemini profile, hit Save, and the agent on the web updates automatically—no extra deploy step for that choice.
Endpoints as tools
One of the features that stands out in DYPAI: your own endpoints can become tools for the agent. Endpoints you already use from the UI—for create task, list tasks, and so on—can be turned into tools and assigned to the agent.
You can do that manually (for example turn create task into a tool and see it under Tools in the assistant), or you can ask your AI to wire create, search, group, and delete tasks so everything is connected. Send the prompt and wait until implementation finishes.
Verify in DYPAI Web
Refresh DYPAI Web: you should see those endpoints enabled as tools for the agent—so you can inspect what the AI configured.
End-to-end test
Back in the chat, ask the assistant to create a task in a test group. It should run the tools it has; you should see a success message. Refresh the app, open the test group, and confirm the task the agent created. In the dashboard, the same flows appear as tools tied to your agent.
Why this scales
As you grow your backend, agents inside DYPAI can keep using the same tools you built for users. That opens the door beyond a single visual app: other products and AI systems can connect to DYPAI and run those tools too.
What you'll learn
- Choosing between DYPAI Managed (no keys, billed as AI Credits) and BYOK credentials
- Implementing an assistant endpoint with the
agentnode and choosing the model - Turning endpoints into tools (
is_tool) and assigning them to the agent - Testing task creation and confirming tools in DYPAI Web
Generate images too
Beyond chat, the Image Generation node lets an agent or endpoint create images. It's managed-only (no BYOK), billed as AI Credits, and returns a storage URL you can serve or save.
Next
Continue with Payments. More features coming in future videos.