llm-gateway

Section: section6-llm-infrastructure Subsystem: telegram-services_v2 Port: 3515 Deploy type: image-copy Health: http://llm-gateway:3515/health Compose: telegram-services_v2/phase2-section6-llm-infrastructure/docker-compose.yml
Fragile: CRITICAL FRAGILE: ipv4_address MUST be pinned to 172.28.0.74 in compose. The iptables ACCEPT rule in prysm-llm-gw-enforce.sh must match this IP exactly. After any power outage or container recreation, verify with `iptables -L DOCKER-USER -n`. IP drift causes silent traffic DROP in DOCKER-USER chain — LLM jobs will timeout with no application-level error for up to 18+ hours before detection. IMAGE-COPY — requires build + up -d.
Description

LLM routing gateway — routes inference requests to appropriate LLM backends

Role

Receives LLM requests from agent-brain, routes to cloud or local (ollama) backends; enforces rate limits

Dependencies

None configured

Runbooks

LLM Gateway — Diagnose and fix iptables IP drift network-routing

Behavioral Assertions

None

Diagnostic Tools

None

Recent Issues

#TitleSeverityStatusCreated
#651[anthropic] BILLING_LIMITcriticalopen6/1/2026
#312[anthropic] BILLING_LIMITcriticalresolved5/8/2026
#236[anthropic] BILLING_LIMITcriticalresolved5/6/2026
#232[anthropic] BILLING_LIMITcriticalresolved5/6/2026
#99[anthropic] BILLING_LIMITcriticalresolved4/27/2026
#98[anthropic] BILLING_LIMITcriticalresolved4/27/2026
#93[anthropic] BILLING_LIMITcriticalresolved4/27/2026
#80[anthropic] BILLING_LIMITcriticalresolved4/27/2026
#79[anthropic] BILLING_LIMITcriticalresolved4/27/2026
#1LLM Gateway iptables IP drift — silent traffic DROP for ~18 hourscriticalresolved4/19/2026