Multi-Protocol Gateway Runbook

Set to off / 0 / false to disable a protocol without redeploying binaries.

Multi-Protocol Gateway Runbook

Feature flags

Env	Default	Effect
`GATEWAY_INBOUND_CLAUDE`	on	Mount `POST /v1/messages`
`GATEWAY_INBOUND_GEMINI`	on	Mount Gemini generateContent routes
`GATEWAY_INBOUND_RESPONSES`	on	Mount `POST /v1/responses` (minimal set)
`GATEWAY_RELAY_THINKING_TO_CONTENT`	off	`separate` / `merge` thinking stream modes

Set to off / 0 / false to disable a protocol without redeploying binaries.

Prerequisites

Channel relay enabled (GATEWAY_CHANNEL_REGISTRY=on + channels configured)
PAT or JWT with workspace:chat scope
Upstream channels with correct providerType (openai, claude, gemini)

Smoke checks

bash

1# Claude Messages inbound -> OpenAI upstream (replace token/model)
2curl -sS -H "Authorization: Bearer $AGX_PAT" -H "Content-Type: application/json" \
3  -d '{"model":"deepseek-chat","max_tokens":64,"stream":false,"messages":[{"role":"user","content":"ping"}]}' \
4  http://127.0.0.1:8088/v1/messages
5
6# Gemini generateContent
7curl -sS -H "Authorization: Bearer $AGX_PAT" -H "Content-Type: application/json" \
8  -d '{"contents":[{"role":"user","parts":[{"text":"hi"}]}]}' \
9  "http://127.0.0.1:8088/v1beta/models/gemini-1.5-pro:generateContent"

Troubleshooting

Symptom	Check
404 on `/v1/messages`	`GATEWAY_INBOUND_CLAUDE=off` or old binary
500 channel relay required	No active channels / registry off
Empty stream	Upstream SSE format; verify channel `providerType`
Rate limit mapped wrong	See `adaptor/errors_map.go`

Audit fields

Cross-protocol calls add inbound_protocol, outbound_protocol, reasoning_effort, thinking_mode on gateway_audit_events JSON payloads.

Made-with: Damon Li