> ## Documentation Index
> Fetch the complete documentation index at: https://docs.prompt-wall.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Integration modes

> Three ways to integrate PromptWall — Proxy, Verify, Events.

PromptWall supports **three integration modes**. Pick the one that
fits your existing setup:

<CardGroup cols={3}>
  <Card title="Full Control" icon="bolt" href="/modes/full-control">
    Drop-in OpenAI replacement. Customer changes `OPENAI_BASE_URL`,
    keeps their OpenAI key, adds an `X-PromptWall-Api-Key` header.
  </Card>

  <Card title="Verify" icon="shield-check" href="/modes/verify">
    Customer calls their LLM directly, then sends the prompt + answer
    to `/v1/verify` for validation and policy enforcement.
  </Card>

  <Card title="Events" icon="chart-line" href="/modes/events">
    Customer ships traces to `/v1/events` from their own SDK / vendor /
    log import. Observability only — no enforcement.
  </Card>
</CardGroup>

## Decision guide

<AccordionGroup>
  <Accordion title="Pick Proxy when…" icon="bolt">
    * You already use the OpenAI SDK / OpenAI-compatible API.
    * You want zero application code changes (just two env vars).
    * You want runtime control to be **available** (opt-in via header
      or per-tenant policy) without rewriting your app today.
    * You want observability immediately, with no glue code.
  </Accordion>

  <Accordion title="Pick Verify when…" icon="shield-check">
    * Your LLM call is wrapped in custom logic you don't want to move.
    * You already retrieve grounding data and want PromptWall to check
      the answer against it.
    * You want **always-on** runtime control (block / rewrite / regenerate)
      sitting after every LLM call.
  </Accordion>

  <Accordion title="Pick Events when…" icon="chart-line">
    * PromptWall did **not** execute the LLM call (Moveo, Intercom Fin,
      Salesforce Einstein, or your own SDK already in production).
    * You want observability without changing the request path.
    * You're backfilling historical traces from a log archive.
  </Accordion>
</AccordionGroup>

## Required keys per mode

| Mode   | OpenAI API key          | PromptWall API key                                |
| ------ | ----------------------- | ------------------------------------------------- |
| Proxy  | ✓ (forwarded to OpenAI) | ✓ (header `X-PromptWall-Api-Key`)                 |
| Verify | ✗ (you call your LLM)   | ✓ (header `X-API-Key` or `Authorization: Bearer`) |
| Events | ✗                       | ✓ (header `X-API-Key`)                            |

> **The customer's OpenAI billing is unchanged in Proxy mode.** Their
> `Authorization: Bearer sk-...` header is forwarded verbatim — they
> keep paying OpenAI directly. PromptWall's role is purely
> observation + governance.

## Same dashboard, no matter what mode

All three modes write into the same canonical `traces` table and
appear on the unified [Observability dashboard](https://www.prompt-wall.com/observability).
You can filter by:

* `use_case` (e.g. `customer_support`, `coding_assistant`)
* `team_id` (e.g. `support-eu`)
* `app_id`
* `integration_id` (`proxy:openai`, `sdk:my-app@1.0`, `vendor:moveo`, …)

Adding more dimensions never increases your bill — PromptWall prices
on **volume** (requests + tokens), not on dimensions.

→ See [Authentication](/authentication) for full key details, then
pick a mode page for ready-to-paste snippets.