Same runtime, your model bill.

Why pick this over Pro

• You already have an Anthropic or OpenAI contract with a rate card better than what we’d pass through.
• You need to keep model usage on your existing vendor billing (procurement, finance, or compliance reasons).
• You want to use Bedrock or Vertex for data-residency or enterprise inference guarantees.
• You want fine-grained per-team or per-project model attribution in the upstream provider’s dashboard.

If none of those are true: pick Hosted Pro instead. It’s simpler and the price difference assumes light usage.

1 · Sign up

Go to concierto.nstack.ai and create an account the same way the Pro flow does (email + code, or Google).

2 · Pick Concierto Connect

On the plan-choice step of onboarding, pick Concierto Connect. Stripe takes a card on file — your $19 / month subscription is for platform access only; model usage will go to the provider you wire up next.

3 · Add your model keys

After workspace creation, go to Settings → Secrets. Add at least one of:

• ANTHROPIC_API_KEY — Claude (Haiku, Sonnet, Opus). Most agents in Concierto are tuned for Claude; if you only add one key, this is the one.
• OPENAI_API_KEY — GPT-5, GPT-4, embeddings. Required if any of your agents are pinned to an OpenAI model.
• AWS_BEDROCK_* or GOOGLE_VERTEX_* — see Bedrock / Vertex notes below for the full variable set.

Secrets are encrypted at rest and only decrypted in the runtime’s process memory when a task starts. They never appear in the UI after you save them.

4 · Create an agent that uses them

Create an agent (“+ Agent” or pick a Quickstart pack). On the agent’s Credentialstab, pick the model. By default agents inherit the workspace’s first available key — Claude if you only added Anthropic. To pin an agent to a specific provider/model, override on this tab.

How key routing works

1. Agent runs a task. The runtime resolves the model: explicit override on the agent → workspace default → platform fallback (Pro plans only).
2. For BYOK, only the first two steps apply. If neither resolves a key for the model the agent needs, the task fails with a clear missing model key error.
3. Workspace-level model routing (e.g. “all designer agents use Claude Sonnet, all coders use GPT-5”) is configured under Settings → Model routing.

Bedrock / Vertex notes

Both are supported but require more environment than a single key. For AWS Bedrock you need:

AWS_BEDROCK_REGION       = us-east-1   # or your region
AWS_BEDROCK_ACCESS_KEY_ID
AWS_BEDROCK_SECRET_ACCESS_KEY
AWS_BEDROCK_DEFAULT_MODEL = anthropic.claude-3-5-sonnet-20240620-v1:0

For Google Vertex:

GOOGLE_VERTEX_PROJECT_ID
GOOGLE_VERTEX_LOCATION
GOOGLE_VERTEX_CREDENTIALS_JSON   # the service-account key as one string

We treat these as provider profiles on the runtime — a single agent can be pointed at one profile and the rest of the workspace can stay on Anthropic. Useful for piloting Bedrock without forcing a hard cut-over.