DSPy classifier workspace
Use DSPy 3.4.1 from Python or run the included helpdesk-classifier CLI. The CLI keeps training, validation and heldout examples separate, saves a selected few-shot program and independently rereads its results. Accuracy improvement is not guaranteed.
Availability
Package 0.1.0 is currently Limited on AWS Marketplace and is not publicly available for subscription. These pages preview its documented behavior. Production-readiness review remains unresolved. Upstream DSPy 3.4.1 retains its Alpha development-status metadata; no stable-release claim is made here.
- Delivery
- Amazon Machine Image in your own AWS account
Quick start
Full guide →What is on the image¶
- The locked base DSPy 3.4.1 SDK at
/opt/dspy/venv/, usable by your own Python programs. - A synthetic three-label classifier workflow and the
sushi-dspycommand. - An Ubuntu-owned workspace, private provider configuration and saved-result directories.
- A standard-library reader that checks hashes, split isolation and recomputed metrics.
Optional SDK extras and additional serving platforms are not all installed. There is no web UI, application API listener, bundled model, local inference service, provider account or provider key. The classifier uses a remote HTTPS OpenAI-compatible endpoint operated by you or another provider.
Instance and access¶
The documented configuration uses Ubuntu 24.04 x86_64 on t3.medium, at least
4 GiB RAM and 24 GiB encrypted root storage. A GPU is unnecessary for a remote
endpoint. Use Session Manager, keep application inbound ports closed and allow
outbound HTTPS to your chosen endpoint and Systems Manager.
Software scope and costs¶
This is paid repackaged open-source software. Charges cover packaging and the included workspace, rather than a bundled technical-support agreement or hosted inference. The configured offer is USD 0.03 per running instance-hour with a five-day software-price trial for one concurrent instance; it is not a public offer yet. Confirm availability and terms before any future subscription. EC2, EBS, data transfer and endpoint infrastructure or provider usage cost extra, including during a software-price trial. The installed SDK has no trial expiry or paid-feature unlock.
The included CLI has bounded labels and dataset sizes. Those limits do not restrict the signatures or sizes of your separately authored SDK programs. A completed workflow does not establish production suitability or model quality.
Where to go next¶
- Getting started: workspace setup and the classifier walkthrough.
- Configuration: endpoint credentials, custom data and your own SDK code.
- Operations: saved-result readback, backups and support.
- Troubleshooting: configuration, datasets and measured accuracy.
Upstream project¶
DSPy is MIT-licensed by Stanford Future Data Systems. Dependency licenses remain
separate. Original wheels, source archives and notices are retained under
/opt/dspy/ and /usr/share/doc/dspy/. Software Sushi is independent of the
upstream project. Read the DSPy documentation.
Last updated November 16, 2026 · Suggest an improvement