@openbastion-ai/proxy

Core proxy server for Bastion — the open-source AI gateway. This package contains the middleware pipeline, provider implementations, and server runtime.

Install

npm install @openbastion-ai/proxy

Programmatic Usage

import { createServer } from "@openbastion-ai/proxy";

const { app, config } = await createServer("./bastion.yaml");
// Server is now running on the configured port

Architecture

Every request flows through an ordered middleware pipeline:

Rate Limit → Injection → Policy(request) → Cache(request)
  → [Provider] →
Cache(response) → Policy(response) → Audit

Middleware

| Middleware | Purpose | |-----------|---------| | Rate Limiter | Token bucket per IP/agent, LRU eviction at 10K buckets | | Injection Detector | 12 weighted patterns, NFKC normalization, leet-speak decoding | | Policy Engine | Declarative rules with 5 condition types, pre-compiled regexes | | Cache | SHA-256 keys scoped by agent/team/env, structuredClone isolation | | Audit | Pluggable exporters (stdout, file, HTTP) with graceful shutdown |

Providers

| Provider | Endpoint | |----------|----------| | Anthropic | Messages API with prompt caching | | OpenAI | Chat completions with function calling | | Google | Gemini GenerateContent | | AWS Bedrock | InvokeModel with cross-region inference | | Ollama | Local open-source models |

Security

• Timing-safe auth token comparison (crypto.timingSafeEqual)
• Security headers on all responses (nosniff, no-store, DENY)
• Request ID and header validation
• ReDoS protection on policy regexes
• Graceful shutdown with audit log flushing

License

MIT