Overview

Product video
Run two sequential agents with CrewAI 1.15.28 in an isolated Python 3.12 environment on Ubuntu 24.04. The included analyst reads six synthetic sales CSV rows, then the reporter produces structured JSON with a dataset identity, row count and integer totals by region. An independent reader recalculates every field from the persisted CSV and checks the saved result and manifest. Every run creates a fresh dataset identity and preserves its files.
Begin with t3.medium, 4 GiB RAM and 32 GiB encrypted gp3 root storage. Connect using AWS Systems Manager Session Manager or SSH and run sudo -iu ubuntu bash /opt/sushi/docs/demo.sh after cloud-init completes. The starter runs as ubuntu and opens no web listener. The default demo uses Amazon Nova Lite in us-east-1 and makes between two and eight real model calls. It requires model access, HTTPS egress and an EC2 instance profile with bedrock:InvokeModel permission for amazon.nova-lite-v1:0. Temporary credentials come only from EC2 IMDSv2; no provider keys or model weights are included or stored. A clearly labeled offline-fixture self-test runs without a provider and does not establish model availability or quality.
Results persist under /home/ubuntu/crewai-results. The installed guide explains independent readback after reboot, preserved first-boot identity and consistent backups. In version 0.1.3, the analyst has one fixed CSV calculator and the reporter has no tools. CrewAI executes the real calculator through Bedrock native tool use, and independent readback checks the calculator result and both original task outputs. Code execution, delegation, memory, planning, telemetry and hosted tracing remain disabled. Other CrewAI integrations need their own configuration and dependency review. The CrewAI core library and CLI are included; optional tool packages and boto3 extras are not installed.
CrewAI is licensed under MIT. Dependencies retain their own licenses and original notices, delivered under /usr/share/doc/crewai; source archives and installed inventories are also supplied. Software Sushi provides AMI packaging and setup guidance. This independent offering does not imply sponsorship or endorsement by CrewAI or Amazon Web Services.
Software costs USD 0.03 per running instance-hour. EC2, EBS, data transfer and model-provider usage are billed separately. Contact info@softwaresushi.com for installation questions. Upstream documentation: https://docs.crewai.com/en/installation . Getting started: https://softwaresushi-marketplace-branding-public.s3.us-east-1.amazonaws.com/docs/crewai/getting-started-24243496533c6d5a9d80f1459e73ca2d17e0d37bf820ff197e91497b6be572ab.html
Highlights
- Run a ready-made two-agent CrewAI workflow (analyst and reporter) that turns sample sales data into structured JSON.
- Check results independently: a separate reader recomputes every total from the saved CSV, including after a reboot.
- Uses Amazon Bedrock (Nova Lite) through the instance role, so no API keys are stored; an offline self-test runs without a model.
Details
Features and programs
Financing for AWS Marketplace purchases
Pricing
Free trial
Dimension | Cost/hour |
|---|---|
t3.medium Recommended | $0.03 |
Dimensions summary
Top-of-mind questions for buyers
Vendor refund policy
We offer a full refund of software charges for the first 48 hours billing starts if the product does not perform as described. To request a refund, email info@softwaresushi.com with your EC2 instance ID and a description of the issue. AWS infrastructure costs (EC2, EBS, data transfer) are not eligible for refund by us.
How can we make this page better?
Legal
Vendor terms and conditions
Content disclaimer
Delivery details
64-bit (x86) Amazon Machine Image (AMI)
Amazon Machine Image (AMI)
An AMI is a virtual image that provides the information required to launch an instance. Amazon EC2 (Elastic Compute Cloud) instances are virtual servers on which you can run your applications and workloads, offering varying combinations of CPU, memory, storage, and networking resources. You can launch as many instances from as many different AMIs as you need.
Version release notes
0.1.3: real model runs now use Bedrock native tool use to invoke the fixed CSV calculator through CrewAI. The analyst task must match the actual calculator result; the reporter must preserve its exact integer findings. Saved results include original provider responses, calculator output, the CrewAI after-tool-call receipt and both original task outputs in a version 2 manifest. The independently checked offline fixture remains explicitly labelled. CrewAI 1.15.28 full/core/CLI on Ubuntu 24.04, isolated Python 3.12, preserved workspace identity, source archives and dependency notices. Recommended t3.medium with 4 GiB RAM and at least 24 GiB root storage.
Additional details
Usage instructions
Launch t3.medium with at least 4 GiB RAM and 24 GiB encrypted gp3 root storage. Require IMDSv2. Connect through AWS Systems Manager Session Manager, which initially uses ssm-user, or SSH with port 22 limited to your source address. Wait for cloud-init, then run sudo -iu ubuntu bash /opt/sushi/docs/demo.sh. The starter runs as ubuntu and opens no web listener. Attach an EC2 instance profile that permits bedrock:InvokeModel on arn:aws:bedrock:us-east-1::foundation-model/amazon.nova-lite-v1:0, provides Session Manager permissions if used, and allows HTTPS egress to the relevant AWS endpoints. Enable Amazon Nova Lite access in your AWS account if needed, and keep IMDSv2 reachable by ubuntu. The default demo makes 2 to 8 real model calls in us-east-1 using temporary IMDSv2 role credentials; it reads no local AWS keys and saves no credentials. The analyst invokes the fixed sum_sales_csv calculator once through CrewAI on this run's six-row synthetic CSV; the reporter copies the exact findings into JSON and code adds the grand total from the CSV. A separate reader recomputes the row count and all integer totals from the saved CSV. Seven files are retained under a new /home/ubuntu/crewai-results directory: sales.csv, result.json, provider-trace.json, tool-result.json, tool-step.json, task-outputs.json and manifest.json. Real model runs use native Bedrock tool use and a version 2 manifest; readback checks the original provider responses, actual calculator execution and both task outputs. Failures preserve their task directory. After reboot, rerun sushi-crewai read against the actual result directory with --require-provider. The explicitly labelled sushi-crewai self-test uses an offline fixture without model calls and cannot prove provider access. First boot preserves existing workspaces and the instance identity in /etc/crewai. Back up /home/ubuntu/crewai-workspace, /home/ubuntu/crewai-results and /etc/crewai while tasks are stopped, preserving ownership and modes; restore to the same application version and rerun readback. Read https://softwaresushi-marketplace-branding-public.s3.us-east-1.amazonaws.com/docs/crewai/getting-started-24243496533c6d5a9d80f1459e73ca2d17e0d37bf820ff197e91497b6be572ab.html and /opt/sushi/docs/README.md. Software costs USD 0.03 per running instance-hour; EC2, EBS, data transfer and model-provider usage are billed separately. Installation support: info@softwaresushi.com .
Resources
Vendor resources
Support
Vendor support
Software Sushi provides AMI packaging and setup guidance. Contact info@softwaresushi.com for installation questions. Support coverage requires a separate agreement; this offer does not include upstream project support.
AWS infrastructure support
AWS Support is a one-on-one, fast-response support channel that is staffed 24x7x365 with experienced and technical support engineers. The service helps customers of all sizes and technical abilities to successfully utilize the products and features provided by Amazon Web Services.