AWS Bedrock
Enterprise access through your AWS account
Amazon Bedrock puts frontier and open-weight models inside your own AWS account: your existing IAM policies, your billing, your VPC, your compliance story. If your organization already lives on AWS, the Amazon Bedrock provider entry is usually the shortest path to approved model access — no new vendor, no new invoice.
Before configuring Nx IDE, request access to the models you want in the AWS Bedrock console (model access is granted per account and region).
Authentication
Section titled “Authentication”Open Settings → Providers, choose Amazon Bedrock, and pick one of three options under Authentication Method:
| Method | Fields | Use when |
|---|---|---|
| AWS Credentials | AWS Access Key, AWS Secret Key, optional AWS Session Token | You have IAM user keys or temporary STS credentials |
| AWS Profile | AWS Profile Name | You use ~/.aws/credentials / SSO profiles on your machine (leave blank for the default profile) |
| Amazon Bedrock API Key | API key | Your account uses Bedrock’s native API keys |
All secrets are stored in the editor’s Secret Storage. Whichever method you choose, the principal needs IAM permission to invoke Bedrock models (for example bedrock:InvokeModelWithResponseStream).
Then set the AWS Region from the dropdown — it lists the Bedrock regions from us-east-1 through eu-central-1 to us-gov-west-1. Model availability differs per region, so pick the region where your model access was granted.
Routing options
Section titled “Routing options”- Use cross-region inference — routes requests through AWS inference profiles that can serve your request from a neighboring region for better availability and throughput. Recommended for busy accounts.
- Use Global inference (auto-select optimal AWS Region) — shown for supported Claude models (Sonnet 4/4.5/4.6, Haiku 4.5, Opus 4.5–4.8). Lets AWS pick the optimal region globally; when both are enabled, Global inference takes priority over cross-region.
- Service Tier — for supported models, choose Standard, Flex (lower price, higher latency), or Priority (lower latency, higher price).
- Use custom VPC endpoint — keep traffic on your private network by entering a VPC endpoint URL, e.g.:
https://vpce-xxx.bedrock.us-east-1.vpce.amazonaws.com/https://gateway.my-company.com/route/app/bedrockPrompt caching
Section titled “Prompt caching”When the selected model supports caching, an Enable prompt caching checkbox appears. It is on by default — Nexgile Code marks stable prompt sections as cache points so repeated context bills at cache-read rates instead of full input rates. Turn it off only if your account or region rejects cache fields. This explicit toggle is a Bedrock-specific control; most other providers cache automatically with no switch. See Model Behavior & Costs.
Extended context
Section titled “Extended context”For Claude Sonnet 4, Sonnet 4.5, Sonnet 4.6, and Opus 4.6, an Enable 1M context window (Beta) checkbox extends the context from 200K to 1M tokens (tiered pricing applies). Opus 4.8 and 4.7 don’t need the toggle — their 1M context is built in.
Models available through Bedrock
Section titled “Models available through Bedrock”The default model is anthropic.claude-opus-4-8. Families in the catalog (verify the exact roster in your region’s dropdown):
| Family | Notes |
|---|---|
| Anthropic Claude | Opus 4.x, Sonnet 4.x and 3.x, Haiku 4.5/3.5 — the main agentic-coding line |
| Amazon Nova | Pro, Lite, Micro, Nova 2 Lite — AWS’s own fast/cheap tiers |
| Meta Llama | Llama 3.x instruct models, 1B through 405B |
| OpenAI open-weight | gpt-oss-20b, gpt-oss-120b |
| Amazon Titan | Legacy text models |
Bedrock’s own catalog is broader than this. Nx IDE is the TAA-compliance edition, so Chinese-origin model families are filtered out of the dropdown even when your AWS account has access to them — see Providers Overview.
Custom ARN
Section titled “Custom ARN”If your organization uses provisioned throughput, imported models, or an application inference profile, select custom-arn at the bottom of the model list and paste the full ARN, for example:
arn:aws:bedrock:us-east-1:123456789012:foundation-model/my-modelThe panel validates the ARN format and warns if its region doesn’t match your selected region.
Practical notes
Section titled “Practical notes”- Claude on Bedrock behaves like Claude from Anthropic directly — same thinking controls, same image support — with pricing set by AWS. See Anthropic for model guidance.
- Throughput on a fresh account can be modest; request quota increases (or enable cross-region/global inference) if you see throttling on long agent sessions.
- Costs land on your AWS bill; the live per-task cost readout in Nx IDE still tracks usage as you go, and you can cap it — see Command Guardrails & Limits.