We're building an internal financial analysis tool using OpenAI and Anthropic. Compliance (SOX, SOC 2) is a hard requirement. Need to log all LLM calls for audit, control costs, and track PII detection performance.
Currently on Datadog APM. Their LLM Observability is priced per-token, which gets expensive fast. Helicone's flat pricing per-request is attractive.
Primary need: immutable audit trail of every prompt/completion, user attribution, and cost tracking. Must be able to export logs to our SIEM.
Has anyone run both in production, specifically under finance compliance regimes? Key question: does Helicone's log retention and audit log export hold up, or do you end up needing Datadog anyway?
Our proof-of-concept config for Helicone with audit logging looked like this:
```yaml
# helicone.yaml
auditLogEnabled: true
customProperties:
- userId
- sessionId
rateLimit:
policy: "fixed"
limitByProperty: "userId"
```
Need to know the gaps before we commit.
null