Data at rest

Where Context stores data, which component accesses each store, and who decides its encryption and retention.

Where data is stored

flowchart LR
  subgraph aws["AWS services you manage"]
    rds["Amazon RDS for PostgreSQL<br/>users, roles and grants, tasks, agents, connections"]
    drive["Amazon S3, drive bucket<br/>drive files, skills, task outputs and transcripts"]
    logs["Amazon S3, logs and traces<br/>logs kept 7 days by default; traces"]
    ch["Amazon S3, ClickHouse data<br/>traces of model calls, including their content"]
  end
  subgraph eks["Inside the Amazon EKS cluster"]
    chv["ClickHouse volume<br/>metadata and cache"]
    nats["NATS stream volume<br/>run event log for live output"]
    eph["Sandbox ephemeral volume<br/>working files of one run; deleted with the Pod"]
    sec["Kubernetes Secrets<br/>database password, signing keys, provider keys"]
  end
  app1["Application tier<br/>workers, Temporal, OnlyOffice"] --> rds
  app2["Application tier<br/>drive gateway, presigned links"] --> drive
  lt["Loki, Tempo"] --> logs
  chs["ClickHouse"] --> ch
  chs --> chv
  st["Streaming service"] --> nats
  sb["Agent sandbox<br/>one per task run"] --> eph
  inst["Installer and application"] --> sec

Stores

Store	Contents	Access	Protection at rest	Retention
Amazon RDS for PostgreSQL	Users, organizations, workspaces, roles, grants, tasks and run records, agent definitions, connector connections, sandbox identity records	Application tier and workers; Temporal; OnlyOffice	Your AWS configuration	Your backups
Amazon S3, drive bucket	Drive files, skills, task outputs and task transcripts	Application tier; drive gateway; presigned links issued by the application tier	Your AWS configuration	Until deleted by users; your backups
Amazon S3, log and trace storage	Application and sandbox logs; traces	Loki and Tempo through their own IAM roles	Your AWS configuration	Logs: Loki retention period (default 168h, seven days); traces: your configuration
ClickHouse	Traces of model calls, including their content; data in Amazon S3, metadata and cache on a cluster volume	Application tier, through ClickHouse	Your configuration for Amazon S3; cluster volume encryption	Your configuration
NATS stream	Run event log for live output	Streaming service	Cluster volume encryption	Your configuration
Kubernetes Secrets and installer configuration	Application secrets (database password, signing keys, provider keys)	Application components; installer	EKS secret encryption	Your configuration
Sandbox ephemeral volume	Working files of one task run	That sandbox only	Node disk	Deleted with the Pod

Logs can include prompt text and tool output. Isolation between organizations, workspaces and users is enforced by the application tier on every request (see the permissions reference). Stores are shared by the deployment; there are no per-organization encryption keys.

For the retention settings an admin controls in the product, see Data retention.