DYPAI Managed AI
Every DYPAI agent runs in one of two AI modes. DYPAI Managed AI is the zero-config default: no API key, no provider account — DYPAI routes the request through its managed (OpenRouter-backed) catalog and meters the usage as AI Credits. BYOK is the alternative, where you bring your own OpenAI / Anthropic / Gemini key and pay the provider directly.
The happy path needs no credential
Set the agent's provider to DYPAI Managed, pick a model, and you're done. No keys to create, no billing accounts to wire up. Building and running agents simply draws from your AI Credit balance.
The two modes at a glance
| DYPAI Managed AI | BYOK (your own key) | |
|---|---|---|
| API key needed | No | Yes — OpenAI / Anthropic / Gemini |
| Where it's set | provider: DYPAI Managed | A credential under Credentials |
| Billing | AI Credits (DYPAI bills you) | You pay the provider directly |
| Markup | Credit conversion (not pass-through) | None — DYPAI doesn't mark up BYOK tokens |
| Model catalog | Plan-gated, managed org-level | Anything your key supports |
| Setup effort | Zero | Get a key, save a credential |
How Managed billing works
When an agent runs in Managed mode, its token usage is metered as AI Credits — the same unit that powers building and editing your app. Pricing is expressed as AI Credits / 1M tokens, so cheaper, faster models cost fewer credits per request than larger ones.
Alongside the usual usage block, a Managed agent response includes a managed_ai_billing block describing what the call cost:
{
"usage": {
"prompt_tokens": 1820,
"completion_tokens": 142,
"total_tokens": 1962
},
"managed_ai_billing": {
"model": "google/gemini-2.5-flash-lite",
"credits_charged": 0.42,
"input_tokens": 1820,
"output_tokens": 142
}
}
It's a credit conversion, not pass-through
Managed pricing applies a credit conversion on top of the underlying generation cost — it is not pass-through and it is not "no markup". The "no markup" promise only applies to BYOK, where DYPAI never touches your provider bill.
Credits are drawn from your monthly plan credits first, then from any purchased top-up packs. Free plans get +5 AI Credits/day; Pro gets 100/month (plus 100 one-time launch credits on activation). Your deployed apps stay live regardless of credit balance — credits only meter AI build/edit and Managed inference work. See AI Credits for the full model.
Picking a model (and what's gated)
On DYPAI Managed AI, which models you can pick is plan-gated. The catalog is managed at the organization level on the AI Models page at /{org}/ai-models, so an org owner controls which managed models are available across all projects in the workspace.
Set the provider to DYPAI Managed
In the agent node, choose DYPAI Managed as the provider. No credential field appears.
Pick a model from the managed catalog
The dropdown only shows models your plan allows. To change what's available, an org owner edits the AI Models page (/{org}/ai-models).
Tune the call
max_output_tokens (default 1024) is clamped by your plan on Managed, so very large requests may be capped automatically. temperature, memory_mode, and tool_ids work the same as in any other mode.
Managed-only nodes
The Image Generation node is Managed-only — there's no BYOK path for it. It's billed as AI Credits (priced from the real generation cost) and returns a storage URL.
When to use BYOK instead
Choose BYOK when you want to use a specific model that isn't in your managed catalog, when you already have provider credits to burn, or when you'd rather keep AI billing on your own provider account. Add an OpenAI / Anthropic / Gemini key under Credentials, set that provider in the agent node, and pick the credential — DYPAI won't mark up those tokens. See Models & Providers and Credentials.