Make Evotown useful for enterprise visibility without collecting unnecessary employee or secret data.
- Collect structured run metadata before raw logs.
- Redact secrets before upload whenever possible.
- Treat artifacts as potentially sensitive.
- Keep source run lineage for every promoted asset.
- Separate observability from employee surveillance in product language and permissions.
- API keys and tokens
- customer data in prompts or artifacts
- source code and private repository paths
- local filesystem paths
- employee identifiers
- model responses containing confidential material
- token-scoped engine authentication
- HMAC or bearer auth for mutating ingest requests
- TLS for production deployments
- artifact size limits
- log excerpt limits
- retention policy for runs, artifacts, and events
- role-based access for enterprise console pages
Connectors should redact before upload. Evotown should still apply server-side redaction as defense in depth.
Examples:
- replace secret-like values with stable fingerprints
- store file hashes instead of raw file content when possible
- keep full artifact bundles optional, not mandatory
Enterprise deployments should support team-level views by default and restrict per-employee drill-down to authorized roles.
Evotown should make visible:
- which policy collected which data
- why a run or artifact was retained
- who reviewed or promoted an asset