General
What is PromptWall, in one sentence?
A governance + observability layer that sits between your app and your LLM, so you can audit every prompt and answer, enforce policies (PII, brand-safety, jailbreak protection), and prove compliance — without re-writing your app.Who is it for?
Companies running LLM features in production who need:- An audit trail of every AI interaction
- Real enforcement (not just logging) on regulated / customer-facing flows
- A vendor-neutral abstraction (works with OpenAI, Anthropic, Bedrock, Azure OpenAI, Google Vertex, anything OpenAI-compatible)
How is this different from observability tools like Langfuse / Helicone / LangSmith?
Those are pure observability (logging, traces, evals). PromptWall does that plus real-time enforcement: it can block, rewrite, or fall back on unsafe answers before they reach your user. The Verify and Full Control modes join the critical path; Events mode is async-only and similar in scope to those tools (at $30/M tokens it’s price- competitive, but the dashboard is governance-flavoured rather than eval-flavoured).Pricing
How much does it cost?
Per-mode token pricing, drawn from a USD-credit pool:- Events: $30 per 1,000,000 tokens
- Verify: $90 per 1,000,000 tokens
- Full Control: $180 per 1,000,000 tokens
Is there a free tier?
Yes — $50 of free credits when you sign up, valid for 30 days, no credit card.What does “1 token” mean?
PromptWall counts the same way OpenAI / Anthropic do —prompt_tokens
completion_tokensfrom the LLM’s reported usage. We trust your reported counts (Events / Verify) or measure ourselves (Full Control).
Can I get a volume discount?
Yes — annual commits over $25,000 get tiered discounts. Email sales@prompt-wall.com.Integration
Which mode should I pick?
Start with Events unless you already know you need enforcement. See Choosing a mode for the full decision tree.Can I use multiple modes at once?
Yes — most production deployments do. Provision one App per mode/environment in Settings → Apps and route traffic to the right key.Does PromptWall support streaming responses?
Yes, on Full Control mode (stream: true). The stream is OpenAI-SSE-
compatible. Verify mode does not stream — it operates on completed
answers.
Does PromptWall support tool / function calling?
Yes. On Full Control, forward the sametools array your LLM SDK
expects. PromptWall scans tool outputs for prompt injection before
re-injecting into the conversation.
Which LLMs are supported?
- OpenAI (gpt-4o, gpt-4o-mini, gpt-4, gpt-3.5)
- Anthropic (Claude 3.5 Sonnet, Haiku, Opus)
- AWS Bedrock (Claude, Llama, Titan)
- Azure OpenAI
- Google Vertex (Gemini)
- Anything OpenAI-compatible (vLLM, Together, Fireworks, Groq, …)
Security & data handling
Where is my data stored?
US-East-1 (Virginia) by default. EU-West-1 (Ireland) available on Enterprise contracts. All data encrypted at rest with AWS KMS.Is PromptWall SOC 2 compliant?
Type II audit in progress (target: Q3 2026). Type I report is available under NDA — email security@prompt-wall.com.How long do you retain prompts and answers?
Default 90 days. Configurable per-org (Enterprise) up to 7 years for compliance use cases. Set in Settings → Data retention.Do you train models on my data?
No. We never use customer prompts or answers for model training. This is contractually guaranteed in our DPA.Do you share data with subprocessors?
Yes — listed at prompt-wall.com/legal/subprocessors. Currently: AWS (hosting + KMS), Render (compute), Cloudflare (CDN + DDoS).What happens to my OpenAI / Anthropic key in BYOK mode?
Encrypted with KMS using a per-org data key. Never logged. Never returned in API responses. Used only to make the LLM call your app authorized.Reliability
What’s the SLA?
Enterprise contracts: 99.9% uptime per calendar month. Below that triggers service credits. Standard plans: best-effort, monitored at status.prompt-wall.com.What happens if PromptWall goes down?
Depends on your mode and your fail-policy:- Events mode: zero impact — calls are fire-and-forget; if we reject them, your user-visible flow keeps working
- Verify mode: your wrapper falls back to either the original LLM answer (fail-open) or a safe fallback (fail-closed) — your choice
- Full Control mode: your wrapper returns a safe-fallback message
Latency overhead?
- Events: zero (async)
- Verify: typically 80–200 ms p95
- Full Control: dominated by the underlying LLM call — PromptWall adds ~50 ms for pre/post-flight scanning
Policies
What’s a policy?
A rule about what’s allowed, defined in Settings → Policies. Each policy has a type (security / privacy / compliance / brand-safety / off-topic / cost), a severity (low / medium / high / critical), and an action (allow / warn / block / rewrite / regenerate / fallback).Does PromptWall come with built-in policies?
Yes — ~25 default policies cover:- Prompt injection / jailbreak attempts
- PII (emails, phones, SSNs, credit cards, names)
- Off-topic responses (configurable)
- Brand safety (toxic language, competitor mentions)
- Compliance (HIPAA, GDPR, financial advice)