Does Cosine store code or PII? Can we opt out of training?
Cosine does not use your data for model training or share it with third parties. Your code, tickets, and documentation always remain private and under your control.
Data handling overview
Section titled “Data handling overview”Code and repository data
Section titled “Code and repository data”When you connect a repository, Cosine securely indexes your code to understand structure and context. This data stays within your chosen deployment (Cloud, VPC, or on-prem) and is never transferred to shared or external environments.
- Data is stored only for the duration of task processing and retrieval.
- Customers can delete repositories or all associated data at any time.
- Enterprise customers can choose to disable all data persistence.
Personally identifiable information (PII)
Section titled “Personally identifiable information (PII)”Cosine does not require or process PII beyond basic account metadata (e.g., email for authentication). Within task data, any incidental PII remains fully contained inside your deployment and is not logged or transmitted externally.
Model training
Section titled “Model training”Cosine never trains shared models on customer data.
- Your proprietary code and documentation are excluded from global training datasets.
- Fine-tuning (if requested) happens only on your instance, within your secure environment.
- When the model is updated, your data does not leave your perimeter.
Data retention and deletion
Section titled “Data retention and deletion”Customers control retention policies:
- Cloud deployments – Data stored in encrypted form and automatically purged after defined intervals.
- VPC/on-prem deployments – Full control over data storage, backup, and deletion.
You can request full deletion of all data artifacts at any time via the admin dashboard or API.
Opting out of training
Section titled “Opting out of training”All customers are opted out of global training by default. No additional configuration is required.
If you request private fine-tuning on your own codebase, it occurs entirely inside your environment — with your approval, custom data governance, and no data egress.
Why this matters
Section titled “Why this matters”Most AI tools rely on shared model training pipelines, which can expose sensitive intellectual property. Cosine’s architecture was built to eliminate that risk, giving you the benefits of AI automation without compromising ownership or privacy.