Full Control
Recommended. Drop-in OpenAI replacement. Change
OPENAI_BASE_URL
to PromptWall and you get observability immediately. Runtime control
is opt-in.Verify
Keep calling your LLM directly. POST the prompt + answer to
/v1/verify and PromptWall validates it against your policies.Events
Observability-only ingest. Use when PromptWall did not run the
LLM (your own SDK, a vendor like Moveo, or a log archive).
How the modes compare
| Feature | Proxy | Verify | Events |
|---|---|---|---|
| Endpoint | POST /v1/chat/completions | POST /v1/verify | POST /v1/events |
| Required keys | OpenAI and PromptWall | PromptWall only | PromptWall only |
| Control modes | observe / guard / verify / full | block / rewrite / regenerate | none (read-only) |
| Code change | Two env vars | Two SDK lines | One HTTP POST |
| Observability | ✓ Always on | ✓ Always on | ✓ Always on |
You can mix modes. Some apps use Proxy while others use Verify or Events. They all write to the same canonical event log and show up on the same Observability dashboard.
What you get out of the box
- Org-wide AI activity sliced by use case · team · app · integration
- Cost & token rollups at 1-minute / 1-hour / 1-day granularity
- Trace drill-through for every request (prompt → guards → upstream → policy hits)
- Failure-cluster panel showing the top patterns this week
- Policy synthesis that watches traces and suggests new guards
- Usage-based billing primitives (volume only — never per use case)