Frequently Asked Questions

Frequently Asked Questions

General

What is AgentWall?

AgentWall is an "Agent Firewall" - a proxy that sits between your AI agents and LLM providers (OpenAI, Anthropic, etc.). It provides:

  • Loop Detection: Stops infinite loops automatically
  • Budget Enforcement: Per-run, daily, monthly limits
  • DLP Protection: Masks sensitive data
  • Run Tracking: See every step of your agent's journey

How is AgentWall different from LiteLLM/Portkey?

Feature LiteLLM/Portkey AgentWall
API Proxy ✅ ✅
Multi-provider ✅ ✅
Run-level tracking ❌ ✅
Loop detection ❌ ✅
Per-run budgets ❌ ✅
Kill switch ❌ ✅

AgentWall tracks entire agent tasks (runs), not just individual requests.

What's the latency overhead?

Less than 10ms for non-streaming requests. Streaming adds <1ms per chunk.

Which LLM providers are supported?

  • OpenAI (GPT-4, GPT-3.5, etc.)
  • OpenRouter (100+ models)
  • Anthropic (coming soon)
  • Google (coming soon)

Setup

How do I get an API key?

  1. Go to agentwall.io/admin
  2. Create an account
  3. Navigate to API Keys → Create New Key
  4. Copy your key (starts with aw-)

Do I need to change my code?

Minimal changes! Just update base_url:

Can I use my own OpenAI key?

Yes! Configure your OpenAI key in the AgentWall dashboard, or use pass-through mode.

Loop Detection

How does loop detection work?

AgentWall tracks prompts within a run and detects:

  1. Exact repetition: Same prompt sent twice
  2. Similar prompts: >85% similarity (Jaccard)
  3. Oscillation: A→B→A→B pattern

When is a loop blocked?

By default, loops are blocked at:

  • 2nd exact repetition
  • 3rd similar prompt
  • 4th oscillation step

Can I disable loop detection?

Yes, in the dashboard under Settings → Loop Detection.

What if I legitimately need to send the same prompt?

Use different run_id values for independent tasks:

Budget & Costs

How is cost calculated?

Based on token usage and model pricing:

Model Input (per 1K) Output (per 1K)
gpt-4 $0.03 $0.06
gpt-4-turbo $0.01 $0.03
gpt-3.5-turbo $0.0005 $0.0015

What happens when budget is exceeded?

The request is blocked with HTTP 429 and error type budget_exceeded. The run is killed.

Can I set different budgets for different agents?

Yes! Create multiple API keys with different budget policies.

DLP & Security

What data is detected?

  • API keys (OpenAI, AWS, Stripe, GitHub, etc.)
  • Credit card numbers
  • Email addresses
  • Phone numbers
  • SSNs
  • Private IPs
  • JWTs

Is my data logged?

By default, prompts and responses are logged for debugging. You can:

  • Enable "zero retention" mode
  • Self-host AgentWall
  • Use shadow logging (hashed only)

Is AgentWall SOC 2 compliant?

We're working on SOC 2 certification. Contact us for enterprise security requirements.

Troubleshooting

"Invalid API Key" error

  • Verify key starts with aw-
  • Check key is active in dashboard
  • Ensure no extra whitespace

High latency

AgentWall adds <10ms. If you see high latency:

  • Check your network
  • LLM provider may be slow
  • Use streaming for better UX

Loop detected unexpectedly

Check if you're:

  • Reusing the same run_id across tasks
  • Sending very similar prompts
  • Using retry logic without variation

Budget exceeded too quickly

  • Check model pricing (GPT-4 is 60x more expensive than GPT-3.5)
  • Review token usage in dashboard
  • Increase budget limits

Pricing

Is there a free tier?

Yes! Free tier includes:

  • 100 requests/day
  • Basic loop detection
  • 7-day log retention

What's included in Pro?

  • 10,000 requests/day
  • Advanced loop detection
  • 30-day log retention
  • Slack alerts
  • Priority support

Enterprise pricing?

Contact us at enterprise@agentwall.io for:

  • Unlimited requests
  • Self-hosting
  • Custom SLAs
  • Dedicated support

Support

How do I get help?

How do I report a bug?

Email bugs@agentwall.io with:

  • Your run_id
  • Error message
  • Steps to reproduce

Still have questions? Contact us