The gateway. 50 middleware modules in a chain, every one toggleable at runtime via API.
fallbackrouterProvider failover
modelswitchModel aliasing
regionrouteGeo-based routing
localsyncLocal model sync
abrouterA/B traffic splits
cachelayerResponse caching
embedcacheEmbedding cache
semanticcacheSemantic similarity
costcapSpending limits
tierdropAuto-downgrade
idlekillKill idle requests
outputcapMax output tokens
usagepulseUsage reporting
rateshieldRate limiting
promptslimPrompt compression
tokentrimToken trimming
contextpackContext packing
chatmemConversation memory
langbridgeLanguage detection
voicebridgeAudio transcription
structuredshieldJSON validation
evalgateQuality gating
codefenceCode validation
promptguardInjection detection
toxicfilterContent moderation
guardrailTopic fencing
agegateAge-appropriate
hallucicheckHallucination detect
secretscanSecret detection
agentguardAgent safety
anthrofitClaude via OpenAI SDK
geminishimGemini via OpenAI SDK
streamsnapStream capture
imageproxyVision/image proxy
tenantwallMulti-tenant isolation
ipfenceIP allowlist
keypoolAPI key rotation
llmtapRequest logging
tracelinkDistributed tracing
alertpulseReal-time alerts
driftwatchModel drift detection
compliancelogAudit logging
feedbackloopUser feedback
promptpadPrompt management
promptlintPrompt validation
approvalgateHuman approval
batchqueueBatch processing
multicallParallel fan-out
mockllmDeterministic mock
devproxyDev/staging proxy