Platform overview
Open models, deliberate customization and control over deployment.
This is the current product direction. The capabilities below are still in development.
Model research#
We’re exploring ways to understand and adapt open-weight model behavior, including abliteration and tools such as Heretic on compatible models. Heretic is an independent tool, not a KILL Labs model or an announced partner. Changes need to be measured and documented alongside the capabilities they preserve.
Model customization#
Fine-tuning is part of the planned platform: adapting a model to a domain, tone or workflow. Training formats and configuration details are not finalized yet.
Private inference#
We plan to serve selected models through infrastructure operated by KILL Labs, with developer API access and a customer dashboard. These services are not available yet. Infrastructure requirements, data-handling policies and supported environments will be documented before access opens.
Future model downloads#
Publishing selected artifacts on Hugging Face is a future distribution goal. There are no KILL Labs model download links at this stage. Hosted API availability and downloadable releases will have their own documented status.
Evaluation first#
- Start with a defined use case and an inspectable model.
- Document modifications and evaluate useful behavior and regressions.
- Release a workflow once its requirements and limitations are understood.