Solutions / LLM Infrastructure & Governance / Distillery
No-code fine-tuning, one click deploy
Describe a task in plain English, curate a dataset, and get a right-sized fine-tuned model — trained, deployed, and API-key-secured with one click. Built on clean-architecture Go with zero external dependencies, shipped as a single binary. Includes batch inference, version rollback, and a portable “run anywhere” export so you’re never locked in.
What it does
Plain-English task setup
Describe what you want the model to do — no training config, no hyperparameter tuning required to get started.
Right-sized models
Fine-tunes a model sized to the task rather than defaulting to the largest option, keeping inference cost and latency down.
One-click deploy
Trained models are deployed and secured behind an API key automatically — no separate deployment step.
Version rollback
Every fine-tune is versioned. Roll back instantly if a new version underperforms in production.
Batch inference
Run large batches of inputs through a fine-tuned model without standing up separate batch infrastructure.
Run-anywhere export
Export a trained model in a portable format so you’re not locked into Distillery’s runtime long-term.
How it fits your stack
Route the resulting API endpoint through Chokepoint if you want the same governance and cost tracking you’d apply to a hosted provider.
Get started
# download and run the binary ./distillery serve --port 8081 # kick off a fine-tune from the CLI distillery train --task "classify support tickets by urgency" --data ./tickets.jsonl
Related solutions
Chokepoint
Add PII redaction, policy rules, and cost tracking in front of your deployed Distillery model.
Just247Pipes
Drop your fine-tuned model into an automated workflow as a single pipeline step.