the gap we close
You already run Claude. Your bill doesn't run through AWS.
Your team built on the Anthropic API and it works. Then the requirements arrive. Security wants IAM, PrivateLink, and CloudTrail. Finance wants GenAI spend on the AWS bill, drawing down commitments. Procurement wants one less vendor.
The migration looks like an endpoint swap. It isn't. Auth moves from API keys to IAM. Model IDs change. Quotas work differently. Prompt caching and batch have their own syntax on Bedrock.
We pair you with engineers who have done this before. We map your workloads, run side-by-side evals to prove output parity, and cut traffic over in stages. No regressions, no surprises on the first bill.
From Anthropic API to Bedrock in two to four weeks
Three phases, scoped to your workloads. Most teams cut production traffic over within a month of kickoff.
Assess
Inventory every Claude workload. Map models, features, and traffic to Bedrock equivalents. Plan quotas, regions, and residency, and build the before-and-after cost model.
Migrate
Swap auth to IAM, move to the Bedrock runtime or the Anthropic SDK's Bedrock client. Run side-by-side evals, then cut over in stages with rollback ready.
Optimize
Turn on batch inference at half the on-demand price, prompt caching, and cross-region inference. Wire spend into DoiT Cloud Intelligence so every team sees costs.
meet our team
Engineers who work inside your codebase
Unlike support teams that respond to tickets or consulting firms that deliver a report and leave, Forward Deployed Engineers work inside your codebase.
They know your architecture, your prompts, and your constraints before the first traffic shift. They've done this migration before, and they document everything so your team can repeat it without us.
What you walk away with
Every engagement is scoped to your workloads. Here's what teams typically leave with.
Migration runbook
Endpoint, auth, SDK, and model ID changes documented per service. Your team can repeat it for the next workload without us.
Output parity report
Side-by-side evals of your production prompts on the Anthropic API and Bedrock. You see the evidence before any traffic moves.
Quota and throughput plan
On-demand versus provisioned, cross-region inference profiles, and quota increases filed and approved before cutover day.
Security baseline
IAM policies, PrivateLink, KMS encryption, CloudTrail logging, and Guardrails configured to your security team's requirements.
Cost model and FinOps setup
Before-and-after unit costs, commitment drawdown math, and DoiT Cloud Intelligence dashboards attributing Claude spend to teams and features.
Cutover and rollback plan
Staged traffic shifting with a tested rollback path. If anything looks off, you're back on the previous setup in minutes.
Ready to move Claude to Bedrock?
Same models your team ships with. Billing and security where your cloud lives.
for AWS teams
AWS reps: send us your Anthropic API customers
You have customers running Claude outside AWS. That spend never touches Bedrock and never counts toward their commitment. The customer wants to keep Claude, so nobody moves.
This accelerator is the move. The customer keeps the exact models they run today. The workload lands on Bedrock, the spend draws down their EDP or PPA, and DoiT engineers do the migration work. Nothing lands on your SA's plate.
the payoff
Outcomes you can measure
A migration is done when the numbers say so. Your Claude spend lands on the AWS bill, parity is proven in evals before cutover, and your cost per token drops below what you paid before the move.
Spend on your AWS bill
Claude usage draws down your EDP or PPA and can consume AWS credits, instead of running through a separate vendor invoice.
Zero-regression cutover
Parity evals run before traffic shifts. Your outputs on Bedrock match your outputs today, and you have the report to prove it.
AWS-native security
IAM, PrivateLink, KMS, and CloudTrail apply to your GenAI traffic the same way they apply to everything else you run.
Lower unit cost
Batch inference at half the on-demand price, prompt caching, and model tiering. Most teams pay less per token after the move.
Frequently asked
questions
Are the models different on Bedrock?
No. Bedrock serves the same Claude models Anthropic serves through its own API. Your prompts carry over as they are.
Will I lose features?
Most Anthropic API features have direct Bedrock equivalents, including tool use, streaming, prompt caching, batch, and extended thinking. We map your specific feature usage during Assess and flag any gap before you commit to the migration.
How long does it take?
Two to four weeks for most teams, from kickoff to production traffic on Bedrock. Complex multi-service estates take longer, and we tell you that in the first week, not the last.
Do I have to rewrite my integration?
The changes are contained: auth, endpoint, model IDs, and a few request-shape differences. If you use the Anthropic SDK, its Bedrock client keeps most of your code intact. The runbook documents every change we make.
What happens to my existing Anthropic agreement?
We plan the cutover around your current contract and commitment dates so you're not paying for capacity in two places.
What does it cost?
For DoiT customers, nothing beyond your Bedrock usage. The accelerator is included in the DoiT relationship.
Claude on Bedrock Accelerator
Keep Claude. Change where it runs.
Book a discovery call and we'll scope your migration in the first session.
