AI Agent - Aug 13, 2026

Daybreak Bedrock Responses API and Mantle Setup Guide

Quick Answer

OpenAI says approved Daybreak customers can use Amazon Bedrock’s console or the Responses API through the bedrock-mantle endpoint. For supported local ChatGPT Work and Codex surfaces, configure the amazon-bedrock provider. Requests then go from the local client to Bedrock using AWS-native credentials; the OpenAI-hosted Responses API is not in that path.

Do not paste a generic endpoint, model ID, Region, or price into production config from a blog. Confirm each value in the current AWS account and model card.

Architecture Boundary

Approved local client
  → amazon-bedrock provider
  → Bedrock API key or AWS SDK credential chain
  → supported AWS Region
  → Bedrock Mantle Responses API implementation
  → approved OpenAI model

This path is different from calling the OpenAI-hosted API. AWS owns credential validation, IAM permissions, Region availability, quotas, billing, and Bedrock service logs for the Bedrock request path.

Minimal Provider Configuration

OpenAI’s documented local provider setting is:

model_provider = "amazon-bedrock"

The client can use a Bedrock API key or the AWS SDK credential chain. With API-key authentication, OpenAI documents AWS_BEARER_TOKEN_BEDROCK plus an AWS Region. With AWS SDK credentials, prefer the organization’s existing SSO, named profile, or federated identity controls.

Never commit keys or session tokens. Keep browser login, token exchange, caching, and refresh in an approved credential helper or AWS profile.

Preflight Checks

Before the first request, verify:

  • Daybreak approval for the exact model and surface;
  • Red approval separately if Red is selected;
  • the current AWS model identifier from the model card or account;
  • model support in the selected commercial AWS Region;
  • IAM permission for the exact caller;
  • quota, billing owner, and logging destination;
  • the bounded security target and allowed action list.

OpenAI’s current local guide says Bedrock Mantle endpoints in AWS GovCloud Regions are not supported by those local ChatGPT Work and Codex surfaces. Recheck this boundary before designing a regulated deployment.

Verification Sequence

  1. Start a new local session after changing provider credentials.
  2. Confirm the client reports amazon-bedrock as the provider.
  3. Use a small, non-sensitive, read-only prompt against an approved model.
  4. Confirm the AWS identity, Region, request record, latency, and billing attribution.
  5. Test an intentional permission denial to prove least privilege.
  6. Test expiration, retry, and duplicate prevention without a consequential action.
  7. Add tool execution only after the plain model path and review controls pass.

The enrollment and access guide covers failures before this runtime stage.

Feature Boundaries

An OpenAI-compatible Responses API shape does not make the hosting paths identical. OpenAI documents some local features as unavailable or limited with Bedrock, including Fast Mode and multiple cloud-hosted integrations. Validate every tool, plugin, browser, remote-control, or cloud-dependent feature on the exact supported surface before including it in an operating plan.

For sensitive cyber work, keep the prompt, retrieved data, tools, outputs, and action approvals in the same reviewed data-flow diagram. API compatibility does not automatically transfer data controls or action authority.

Frequently Asked Questions

Does Daybreak on Bedrock call the OpenAI-hosted Responses API?

Not in the local provider path documented by OpenAI. The local client sends requests to Bedrock, which provides an OpenAI-compatible Responses API implementation for supported models.

Which model ID should I use for Daybreak on AWS?

Use the exact identifier shown for the approved model in the current AWS account or model card. IDs and Region availability can change, so do not infer one from a product label.

Can I authenticate this provider with OPENAI_API_KEY?

No. OpenAI documents a Bedrock API key or the AWS SDK credential chain for the amazon-bedrock provider, not ChatGPT sign-in or OPENAI_API_KEY.

Does OpenAI-compatible mean every OpenAI feature is available?

No. Request compatibility does not guarantee feature parity. OpenAI documents several hosted and cloud-dependent features as unavailable or limited on the Bedrock provider path.

Are local Bedrock Mantle clients supported in AWS GovCloud Regions?

OpenAI’s current local setup guide says the covered ChatGPT Work and Codex surfaces do not support Bedrock Mantle endpoints in AWS GovCloud Regions.

Official Sources

Next Decision