SBproxy manages traffic in both directions: the models your apps call and the agents that call your APIs. Govern, transform, and secure every request, serve your own models, and run it all in your own perimeter, from one config.
The full data plane is open source and runs entirely inside your perimeter, from a single binary. No control plane to operate, no data leaving your network.
Web Bot Auth verifies signed AI crawlers and agents using RFC 9421 HTTP Message Signatures with Ed25519 and HMAC. Virtual keys gate partner and customer APIs. mTLS, OAuth2 with DPoP and PKCE, and JWT with JWKS round out the auth surface.
Virtual keys are a runtime resource, not a line of YAML. Mint, rotate, and revoke them through the admin API with no reload, and a revoke takes effect on the next request. Keys are hashed at rest with HMAC-SHA256 and a server pepper, and upstream provider credentials are sealed with AES-256-GCM. Run more than one replica and a gossip mesh keeps the keys, budgets, and rate counters coherent across all of them, so the fleet coordinates itself without an external Redis.
AI agents pay tokens for HTML they did not need. Negotiate Markdown on demand, project JSON, validate against an OpenAPI spec, and emit a live OpenAPI document for buyers. Content negotiation is a transform, not a website rewrite.
Every B2B company is being asked whether their product is agent-ready. Most answer by writing an MCP server. SBproxy's OpenAPI-to-MCP bridge turns the project into a config change. Point the gateway at an OpenAPI spec and the agent gets one tool per (path, method) pair, with parameter schemas mapped, auth carried through, and rate limits applied. No new server. No new audit story.
200+ models behind one OpenAI-compatible API. Fallback chains, weighted, latency-aware, and cost-optimized routing. Hierarchical budgets with cost attribution per workspace, team, and customer. PII redaction, prompt-injection v2, secret scanning, and toxicity guardrails on input and output. MCP gateway federates internal tools under one identity model.
Every request carries one correlation id across all three signals. Prometheus metrics on /metrics, structured JSON logs to stdout or any OTLP collector, and OpenTelemetry traces. AI spans follow the GenAI and OpenInference conventions and carry per-request token counts and derived USD cost, so Phoenix, Langfuse, Grafana, or Datadog render the full LLM trajectory with no custom mapping.
Single binary on a server. Kubernetes operator on a fleet. Air-gapped install on a sovereign network. Same policy graph in every shape. Hot reload with no dropped connections. Bring your own Stripe, SSO, observability, and audit sink.
Single binary on a server. Kubernetes operator on a fleet. Air-gapped install on a regulated network. Same policy graph in every shape.
One static executable. Drop it on a server, point it at a YAML, route real traffic.
CRD-driven control plane. Reconciles policy, certs, and origins across the fleet.
Sealed install. No phone-home, no licence call-out. Suitable for sovereign and defence networks.
API access, content monetization, agent identity, AI routing, and MCP federation in the same YAML. Not five files glued together. Not five vendors stitched in middleware.
# One policy graph: API + content + agents + AI + MCP
origins:
# --- Public content with Pay Per Crawl + Web Bot Auth + Markdown ---
"www.example.com":
action:
type: proxy
url: https://origin.internal:8443
authentication:
type: bot_auth # RFC 9421 signed agents
agents:
- name: openai-gptbot
key_id: openai-2026-01
algorithm: ed25519
public_key: ${OPENAI_BOT_PUBKEY}
policies:
- type: ai_crawl_control # Pay Per Crawl
price: 0.001
currency: USD
header: crawler-payment
transforms:
- type: html_to_markdown # Negotiated on Accept: text/markdown
# --- Partner API with virtual keys + per-customer quotas ---
"api.example.com":
action:
type: proxy
url: https://backend.internal:8443
authentication:
type: virtual_key
policies:
- type: rate_limiting
key: virtual_key
requests_per_minute: 600
- type: waf
# --- Outbound AI gateway with budget + guardrails ---
"ai.example.com":
action:
type: ai_proxy
routing: fallback_chain
providers:
- name: anthropic
api_key: ${ANTHROPIC_API_KEY}
- name: openai
api_key: ${OPENAI_API_KEY}
guardrails:
input: [{type: pii}, {type: injection}]
budget:
limits:
- {scope: team, max_usd: 5000, period: monthly}
# --- MCP gateway federating internal tools ---
"mcp.example.com":
action:
type: mcp
mode: gateway
federated_servers:
- {origin: github.internal, prefix: gh, rbac: read_only}
- {origin: postgres.internal, prefix: db}
authentication:
type: oauth2 # DCR + DPoP + PKCEThe gateway is on GitHub under Apache 2.0 and you can install it right now, so nothing here is gated behind a sales call. Use this form for deployment questions, migration scoping, or anything you would rather not file as a public issue.