Skip to main content
NuBench AI
Get started

Workspace setup

A NuBench workspace is your organisation's governed AI environment: its own organisation record, budgets, policies, and audit trail, running on single-tenant AWS infrastructure.

Create the workspace

  1. Go to www.nubench.ai/signup and complete the self-provision form — organisation name, your name, and your work email.
  2. The workspace is created immediately with a trial organisation and your account as its first organisation admin. No sales call and no credit card are required to start.
  3. Sign in at chat.nubench.ai with the account you just created.

Connect a model provider

NuBench is BYO-first: your workspace calls AI models with your own provider relationship, under your own data-processing terms.

  • Default — Amazon Bedrock. Workspaces start on Amazon Bedrock using NuBench's AWS-native integration. Requests are dispatched with IAM authentication inside the AWS boundary; there is no shared API key.
  • Bring your own key. Register your organisation's OpenAI, Anthropic, or Azure OpenAI key. Keys are stored in AWS Secrets Manager, encrypted with a customer-managed KMS key, and are only read at dispatch time by the governance gateway.

Whichever provider you choose, every request passes through the same governance pipeline — authentication, classification, policy, budget, and audit are provider-independent.

Invite your team

Add users from the admin console and assign each a role — see Roles and permissions. New users sign in through your identity provider once single sign-on is connected, or with NuBench-managed credentials during evaluation.

Set your first policies

Before rolling out broadly, most administrators configure:

  1. A workspace-wide token budget with warning thresholds — see Budget policies.
  2. The DLP mode for prompt scanning (flag, redact, or block) — see Data-loss prevention.
  3. Retention for conversation logs, aligned with your records policy.

What your workspace runs on

  • Single-tenant deployment in AWS (us-east-1), managed entirely as infrastructure-as-code.
  • Aurora PostgreSQL behind RDS Proxy with TLS required; ElastiCache Redis with transit encryption.
  • A dedicated customer-managed KMS key per data store — never shared, never AWS-default.
  • WAF on public endpoints; all workloads inside a VPC.

For the full security picture, see the Trust Center.