Frequently Asked Questions (FAQ)
What is Loopers?
Loopers is an open-source, baremetal, zero-delay firewall for AI agents and LLMs. It operates as a high-performance reverse proxy that sits between your applications/agents and model providers (OpenAI, Anthropic, Gemini, Groq, Bedrock, etc.) or MCP servers. It enforces real-time token budgets, detects and breaks infinite agent loops, and prevents bill shocks like LLMjacking.
How does Loopers stop runaway agent loops?
Loopers features a Loop Detection Engine (v1.1) running three deterministic algorithms:
- Fingerprint Ring: Computes hashes of incoming message signatures, tool parameters, and response structures across sliding windows.
- Velocity Limiter: Detects rapid-fire burst iterations that indicate an unconstrained recursive loop.
- Stall Detector: Identifies non-progressive states where an agent repeatedly issues identical or redundant tool calls without converging on a solution.
When a loop is detected, Loopers severs the connection immediately, returning HTTP status 429 with header X-Loopers-Loop-Detected: true.
How does Loopers compare to LiteLLM and Bifrost?
Loopers is engineered specifically as a baremetal infrastructure-level firewall prioritize speed, zero budget leakage, and agent governance.
| Feature | Loopers | LiteLLM | Bifrost |
|---|---|---|---|
| Language / Architecture | Go (Native Binary) | Python (FastAPI) | Go |
| Peak Throughput | 4,623 req/s | ~176 req/s | ~3,200 req/s |
| Budget Leakage (1k flood) | 0% ($0.00) | 0.17% | 0% |
| P99 Proxy Overhead | 240.98 ms | 46,812.60 ms | ~310 ms |
| Agent Loop Circuit Breaker | Yes | No | No |
| MCP Tool Governance | Yes (Per-tool) | No | Policy-based |
| Storage Security | Zero-Storage Pass-Through | Database Required | In-Memory |
What is MCP Governance in Loopers?
Model Context Protocol (MCP) governance allows enterprise and developer teams to safely expose external tools (databases, shell tools, APIs, web search) to autonomous AI agents. Loopers proxies JSON-RPC 2.0 requests between agents and MCP servers, enabling:
- Per-Tool Cost Controls: Set maximum USD spend per tool call (e.g. max
$0.05per SQL query). - Circuit Breakers: Cap the maximum tool invocations allowed within a single session.
- Blast Radius Prevention: Stop compromised or confused agents from performing lateral movement across multiple sensitive tools.
Is Loopers zero-storage and safe for production data?
Yes. Loopers follows a Zero-Storage Security Model:
- No payload text, completions, or API keys are written to disk or database.
- Request headers and metadata are processed in volatile memory only.
- Budget accounting is recorded atomically in Redis using single-roundtrip Lua scripts without storing prompt content.
How do I install and deploy Loopers?
Loopers can be deployed in under 60 seconds via Docker or binary release:
# Run with Docker
docker run -d \
-p 8080:8080 \
-e REDIS_ADDRESS=localhost:6379 \
tryloopers/loopers:latest
Then point your provider SDK base URL to standard Loopers port:
import openai
client = openai.OpenAI(
base_url="http://localhost:8080/v1",
api_key="your-openai-key"
)