Zero-persistence data processing
ADEQUATA enforces zero-persistence data processing inside the execution pipeline. For each job, Axolotl transfers the files you selected into short-lived compute (AWS Batch containers in us-east-1). Raw files are loaded onto RAM-backed storage, with swap and core dumps disabled. When the job finishes, the container terminates and that storage is destroyed. Accepted results are written back to the output path you chose. Raw dataset contents are not written to disk in the processing environment.
Zero dataset retention means ADEQUATA keeps no copy of your dataset contents after the job. It does not mean that every operational record disappears. Pipeline status, billing, stage summaries, quality metrics, and compliance notes—such as a record that a run was blocked for size—are kept according to our retention policy. A review sample is read from your storage and shown on the review screen. The app does not save that sample as a copy of the dataset.
- When you connect a source, the credentials are stored encrypted in AWS Secrets Manager. During a job they are held in memory only, and they are not written to disk in the processing environment.
- During preflight, Axolotl reads schema and size information from your storage and may pull a small sample so you can resolve column conflicts. That sample is for review, not a stored copy of the dataset.
- During the job, inputs are held in memory and on RAM-backed storage on the job host. Remediation runs against that memory, not against a durable ADEQUATA table and not from disk in the processing environment.
- Write-back copies only the output you accepted to the bucket path you configured. If you skip review, that stage is accepted by default, and the same write-back runs.
Human review before write-back
Human review happens inside the application and is based on a sample. Axolotl does not export your dataset into ADEQUATA storage for that review.
- For each stage you see summary metrics and a sample of rows, and you choose whether to accept or decline.
- If you decline, the input path is left unchanged. The run can finish without promoting that stage’s output.
- If you accept, Axolotl continues to the next stage you selected, or it writes the final result to your output prefix. Skipping review has the same result as accept.
- You can inspect the records of the run: stage summaries, the schema validation artifact, and, if the memory threshold blocked the job, the size-block record.
- Those records describe the process and the billing decision. They are not a durable copy of every row in your dataset.
Data isolation and egress
Your dataset remains in your storage. During a job, a working copy is held in memory on the job host and destroyed when the job ends. ADEQUATA does not keep a copy of the dataset contents, and does not use them to train global models.
| Control | Summary |
|---|---|
| In transit | Browser, application, and cloud storage access use HTTPS (TLS). The app also sends HSTS. |
| At rest (your data) | Remains in your bucket. During a job it is held in memory on the job host and destroyed when the job ends. |
| At rest (secrets) | Encrypted in AWS Secrets Manager. Held in memory only during a job, and not written to disk in the processing environment. |
| Training | Customer content is not used to train or fine-tune ADEQUATA global models |
| App access | Organization-scoped; admin and member roles; SSO on Advanced and Enterprise (SAML/OIDC); Directory Sync (SCIM) on Enterprise |
| Session cookies | On app.adequata.ai, session and sign-in cookies are host-only __Host- cookies. They are httpOnly and encrypted, so page scripts cannot read the tokens. The session idles out after about an hour without activity. |