Model Access & Quotas

Complete these checks in the AWS account that will run Brainbase, in us-east-1. Access in another account, including your organization management account, is not sufficient evidence. This page describes release 2026-09-16.1; always inspect your applied account limits rather than assuming advertised defaults.

Model access is separate from throughput

The release's LLM gateway calls bedrock-mantle.us-east-1.api.aws using an ECS task role and SigV4. It does not require an Anthropic or OpenAI API key. The account must be allowed to use each selected model; a quota increase is not necessarily a model-access approval.

1. Confirm your intended model is in the deployed Brainbase model selector and available to the deployment account through Bedrock Mantle in us-east-1.
2. Have an authorized AWS administrator review applicable model terms, Marketplace model-subscription permissions, payment prerequisites, and organization restrictions. Brainbase subscription and third-party model access are separate.
3. For a new account receiving "model not available for this account" or access-denied errors, contact AWS Support or your AWS account team. Ask whether frontier-model account enablement is required, not only whether a throughput quota is low. This can require an AWS-internal process; do not assume it appears as a trackable Service Quotas request.
4. After installation, run an actual task with the selected model. A model catalog entry or successful test through a different Bedrock endpoint does not prove the appliance inference path works.

Established accounts may already have access; approval is not an extra step everyone must repeat. AWS documents model-access prerequisites. Its Anthropic first-time-use form requirement does not apply to the Mantle path; this does not waive other account eligibility, IAM, or model restrictions.

You do not need approval for every example model to begin. Request what you intend to use. The release also uses anthropic.claude-haiku-4-5 for title generation, so include that auxiliary model when validating the complete experience. The AWS appliance catalog can differ from the hosted SaaS catalog.

Check Bedrock Mantle throughput

In Service Quotas, select Amazon Bedrock and search for Bedrock Mantle. Its per-model input and output token quotas are separate from Bedrock Runtime quotas. Mantle does not expose an RPM quota. Some models have no published per-account quota there; absence is not proof of unlimited capacity or denied access.

For increases, use AWS Support's Amazon Bedrock limit-increase process and specify bedrock-mantle, us-east-1, exact model IDs, and requested input/output tokens per minute. Include expected concurrency, context/output sizes, and measured usage where available. If the quota is not exposed, ask AWS which capacity or enablement process applies. See AWS Mantle quotas.

There is no universal Brainbase TPM minimum: long prompts, tool results, parallel agents, and output reservations change demand. Start with one small task, then measure your representative workload before adding concurrency. A throughput allowance does not guarantee every on-demand request succeeds; transient capacity throttling can still occur. See AWS throughput guidance.

Size Lambda MicroVM capacity

The shipped sandbox image reserves 8192 MiB (8 GiB) per VM. In Service Quotas, select AWS Lambda and locate Max allocated MicroVM memory, quota code L-CD1C0CC4. Use the code to identify it if the display name differs. This is separate from ordinary Lambda function concurrency or per-function memory.

Count both RUNNING and SUSPENDED VMs, including other workloads in the account and region. Suspension reduces compute charges but does not release this capacity. See AWS MicroVM capacity.

These are sizing examples, not AWS defaults or promised approvals. Add other MicroVM workloads and operating headroom. New accounts can have capacity for only one Brainbase VM. Check the applied account quota, request your needed value, and wait for the applied value to change before increasing usage.

aws sts get-caller-identity --region us-east-1

aws service-quotas get-service-quota \
  --service-code lambda --quota-code L-CD1C0CC4 \
  --region us-east-1 \
  --query 'Quota.{Name:QuotaName,Applied:Value,Unit:Unit,Adjustable:Adjustable}'

aws lambda-microvms list-microvms --region us-east-1 \
  --query 'items[].{Id:microvmId,State:state,Image:imageArn,Version:imageVersion}'

Check deployment infrastructure quotas

Review quotas in each service's console or Service Quotas; use the failed resource's error if CloudFormation identifies another account limit. Do not raise every quota blindly. Return to installation after the account and capacity checks.