Quick answer
Use the current WorkSpaces Applications pricing page, not the older launch-post sentence, to estimate agent access. The May announcement still says “no additional cost,” but the live pricing page checked August 12, 2026 documents separate active agent-session fees, a monthly regional free allowance, and a Windows AI-agent user fee. The sources conflict, so the pricing page and AWS calculator should control a purchase decision.
Cost also includes the underlying streaming fleet and may include image builders, stopped capacity, model inference, S3, CloudTrail, CloudWatch, networking, public IPv4, licensed applications, and forwarded services.
Current price components
| Component | Current pricing-page treatment | Verification question |
|---|---|---|
| Agent MCP session | Billed by the second while actively connected; no per-tool-call fee | What is the regional hourly rate and when does idle end the session? |
| Monthly allowance | 10 free agent-session hours per account per Region; reset monthly, no rollover | Does the account and region qualify under current terms? |
| Windows agent user | Flat $6.42 per user per month when an AI agent user connects to a Windows Server fleet | How is the agent user uniquely identified across auth type and Region? |
| Fleet instance | Always-On or On-Demand instance pricing still applies | What instance, capacity policy, running and stopped time are expected? |
| Model | Separate Bedrock or other provider charges | Which model, region, tokens, screenshots, and retries? |
| Evidence and data | S3, CloudTrail, CloudWatch, transfer, and networking may apply | What is captured, retained, queried, and transferred? |
These figures are volatile. Do not copy them into a procurement estimate without reopening the official page.
Resolve the official-source conflict
The launch post was published May 5 and updated July 1 only to mark GA. It still contains an earlier “no additional cost” statement. The live pricing page provides a detailed “WorkSpaces for AI agents fees” section and a worked example that includes both streaming and agent-session charges.
The safe interpretation is not that one source proves a billing error. It is that a dated launch paragraph is insufficient price authority. Capture the pricing page date, selected Region, calculator export, account offer, contract terms, and billing-owner approval.
Current availability
AWS currently lists agent access in:
- US East: N. Virginia and Ohio;
- US West: Oregon;
- Canada: Central;
- Europe: Frankfurt, Ireland, London, and Paris;
- Asia Pacific: Mumbai, Singapore, Sydney, Tokyo, and Seoul.
Availability for WorkSpaces agent access, the selected fleet instance, Amazon Bedrock model, S3 policy, identity provider, and downstream application must all overlap. A service region does not establish model access or data-residency approval.
Current functional limits
The service documentation checked August 12 lists Windows Server images only, Always-On or On-Demand fleets, no VPC endpoints, no multi-session fleets, no Elastic fleets, and one agent connection per unique session. It requires an associated fleet and stack and the latest WorkSpaces Applications Agent.
These limits affect cost. One session per agent changes concurrency planning. An Always-On fleet can accrue capacity charges when no agent is connected. An On-Demand fleet can carry stopped-instance costs. Screenshot retention and verbose logs can expand storage and query spend.
Build a task-level cost model
Estimate the cost of an accepted task, not an hour in isolation:
- forecast concurrent sessions, session duration, retries, idle time, and daily schedule;
- select fleet type, instance, minimum and buffer capacity, and Region;
- add monthly agent users and agent-session time after the applicable allowance;
- add model input, image, output, and retry usage;
- add image builder, application licensing, S3, logs, metrics, network, and public IPv4;
- measure task success and human correction rates in a pilot;
- divide total cost by accepted, reconciled outcomes;
- model the API or browser-automation alternative on the same basis.
Use the desktop vs API vs browser guide for the architecture comparison and the readiness checklist for the budget gate.
Frequently asked questions
Is agent access free?
Do not rely on that claim. The current pricing page lists agent-session fees after a regional allowance and other applicable charges.
What is the free allowance?
The current page states 10 hours per account per Region each month, resetting without rollover. Verify it live.
Which fleets work?
Current documentation supports Always-On and On-Demand Windows fleets and excludes multi-session and Elastic fleets.
Official sources
Source check: August 12, 2026. Recheck live regional rates, free-tier terms, user identity rules, fleet pricing, service limits, model availability, and contract terms before budgeting.