Claude Code

Zero Data Retention

Claude Code ZDR coverage, disabled capabilities, data boundaries, cache misconceptions and Passion8 considerations.

Zero Data Retention is an official mode for eligible Enterprise organizations that avoids normal retention of model-request contents. It does not mean every related record is absent.

ZDR belongs to Anthropic account/organization policy. Separately verify gateway, upstream, CI, MCP and local retention when using Passion8.

#Coverage

The primary scope is model-interaction content such as prompts, completions and tool context. It does not eliminate local transcripts or govern third parties.

Object	ZDR interpretation
Model-request content	Core covered target
Analytics	No prompts/responses, but account/seat/usage metadata remains
Account/organization records	Standard administrative retention
Third-party tools/databases/CI	Outside Anthropic ZDR
Local transcripts	Require local cleanup policy
Passion8 logs	Separate gateway policy

#Capabilities outside the scope

Official documentation distinguishes Code from other product surfaces. Do not assume these are covered:

Scenario	Reason
claude.ai chat	Separate product surface
Cowork	Outside Code ZDR
Third-party integrations	Their own policies apply
Local logs/transcripts	Your filesystem responsibility
Administrative data	Seats, email and settings retained

#Disabled capabilities

ZDR restricts features requiring server-side conversation storage.

Capability	Reason
Code on the web	Server-side session history
Desktop cloud sessions	Persistent session data
Published Artifacts	Hosted page content
Feedback	Sends conversation data
Contributions	Usage metrics only in ZDR organizations

Evaluate alternatives before enabling ZDR if you depend on cloud review, web or Artifacts. Sensitive repositories often stay in local CLI/IDE or self-hosted containers/runners.

#Policy violations and security records

Necessary safety events or metadata may still be retained to detect and handle violations. Therefore:

ZDR does not justify bypassing permissions, audits or DLP.
Sensitive repositories still need rules, hooks, MCP allowlists and network isolation.
Resolve legal/security questions through the enterprise agreement and official support.

#Evaluate Passion8 paths

The actual path is:

Local Claude Code -> Passion8 gateway -> upstream provider

Check each layer:

Layer	Questions
Local device	Transcript lifetime and access to ~/.claude/?
Passion8	Request bodies, results, usage and error logging?
Upstream	Training, retention and ZDR/equivalent eligibility?
CI/CD	Do job logs expose prompts, diffs, secrets or output?
MCP	Are returned data stored, audited or forwarded elsewhere?

Strict retention requires coordinated provider, gateway, MCP, CI and local policies, not merely an environment-variable change.

#Caching versus ZDR

Cache TTL measures inactivity before prefix expiry, not total request retention.

Concept	Meaning
Prompt cache	Reuse prefixes for cost and latency
ZDR	Server-side content-retention policy
Local transcript	Plaintext conversation for resume
Gateway logs	Passion8/custom gateway records

One-hour caching does not promise deletion after one hour; disabling caching does not enable ZDR. See caching and data flows.

#Action checklist

Goal	Recommendation
Apply for official ZDR	Enterprise organization and Anthropic support
Sensitive repositories	Local CLI/IDE with sandbox and deny rules
Cloud capabilities	Identify disabled/degraded features
Local retention	Shorten cleanupPeriodDays and purge when needed
Gateway compliance	Define upstream/gateway retention and logging
Team auditing	Enforce read/export boundaries through managed rules/hooks