BYOK (bring your own keys)

By default, Daslab runs your agents on managed usage: one predictable bill, hard spend caps, prepaid — you never think about provider keys. BYOK is the Enterprise path: connect your own provider accounts, and inference runs on your agreements, billed to you directly.

Supported providers

  • Anthropic — Claude family, all models. Direct API.
  • OpenAI — GPT family, all models. Direct API.
  • OpenRouter — gateway to most other models, including Llama, Mistral, Cohere, etc.
  • Gemini — Google AI Studio.
  • Amazon Bedrock — Claude, Llama, Mistral, Nova, Cohere via your AWS account. Inference stays inside your AWS account.
  • SAP AI Core — models served through your SAP BTP tenant.

On Enterprise, add keys once at the workspace level; every scene in the workspace uses them.

Managed by default; your keys on Enterprise

Pro and Team run fully managed — simpler setup, one bill, and the hard cap means no surprise invoices. Most teams should start there.

BYOK earns its place when you have reasons we can't replicate: negotiated provider agreements, a Bedrock deployment inside your AWS boundary, data terms that require your own contracts, or self-hosting. On Enterprise, connect your keys once at the workspace level and every scene uses your providers directly — we don't mark up your tokens.

Cost visibility

Every job records the model used, the token counts, and — with a connected key — the cost in the provider's currency.

A hard spend cap with a prepaid allowance protects the workspace: hit it and scenes pause until you top up or lift it. No overage invoices. Useful for not waking up to a runaway-cost incident.

Switching models per conversation

You don't pick one model and stick with it. The agent can switch per conversation. Use Claude for planning, GPT for one specific call, Llama via Bedrock for cheap bulk work. The agent loop handles the switch transparently.

We don't lock you into a model family. The "any model, no lock-in" claim is structural, not a marketing line — the integration layer is provider-agnostic.

If you don't have keys

You don't need any: managed usage is the default on Free, Pro, and Team. The free tier includes enough to evaluate; Pro carries real workloads under the hard cap.

If you're an enterprise on Bedrock, you'll want to connect Bedrock — inference stays inside your AWS account and boundary, and the cost model is whatever you've already negotiated with AWS.

What's next

  • Auth — connection model, permission inheritance.
  • Data — residency, retention, no-training.
  • Agents — the model loop.

Updated 2026-04-30