One policy layer.
Every agent.
Outbound LLM traffic already flows through your platform. Now partner agents are calling your APIs and employees are wiring agents into internal data. SBproxy puts inbound and outbound on the same policy graph. Same audit log. Same gateway.
# Outbound AI + MCP federation + cost attribution
origins:
"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}]
output: [{type: secrets}]
budget:
on_exceed: downgrade
limits:
- {scope: team, max_usd: 5000, period: monthly}
- {scope: workspace, max_usd: 50000, period: monthly}
"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: oauth2The problem we hear.
Outbound LLM traffic finally lands under control. Then partner agents start hitting your APIs and employees start piping data into ChatGPT. The control plane fragments across teams. Cost attribution breaks.
AI controls were never going to stay scoped to outbound. SBproxy collapses both directions onto one policy, one audit log, one runtime, so the platform team operates one gateway instead of negotiating across three.
When you put SBproxy in front of your stack.
The work below moves out of the engineering backlog and into a config change.
- Govern outbound LLM traffic and inbound agent traffic on the same gateway.
- Cap AI spend by team, customer, or virtual key. Attribute every dollar.
- Stop prompt injection, PII leaks, and prompt-side credential exposure on every call.
- Make every internal API agent-ready. Without writing an MCP server.
What’s shipping in the data plane.
For engineers in the buying committee. The capabilities behind the outcomes above.
- 01Reach every model from one endpointOpenAI, Anthropic, Google, Bedrock, Azure, Groq, plus self-hosted runtimes. One API. Fallback when a provider stumbles.
- 02Cap AI spend by who is spending itSet budgets per workspace, team, customer, or key. The gateway downgrades to a cheaper model or stops the call. Every dollar attributed.
- 03Federate internal tools under one identityPut GitHub, Postgres, Jira, and your own services behind one MCP catalog. One auth model, one audit log, scoped per tool.
- 04Make any API agent-readyPoint the gateway at an OpenAPI spec and your APIs become MCP tools the agent can call directly. No new server to build.
- 05Stop the leaks before they leavePII redaction, secret scanning, prompt injection defense, and toxicity gates on every call. Block, redact, or flag.
- 06One audit log for every machineInbound agents, outbound LLMs, and MCP tool calls land in the same event stream. Stream to OpenTelemetry, Prometheus, or your warehouse.
- Your platform team is the one on the hook for every agent the business is shipping.
- You already run an AI gateway for outbound. Inbound agent traffic is the new fire.
- Finance keeps asking which team spent what.
Want to talk
before you install?
The 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.