Free-Tier Large Language Model and Image Generation Integration for Local Developer Tools: A Case Study of the Geminigen Stack (free-coding-models, OmniRoute, and Gemini CLI)

This case study documents a free-tier, multi-provider architecture for large language model (LLM) text generation and image generation integrated into local developer tools. The Google image API free tier exhibited a hard limit of zero image quotas, while Groq's free tokens-per-minute (8,000) are smaller than the Gemini CLI system prompt (~18,000 tokens), making route selection decisive. We present the "Geminigen" stack: a free-coding-models (FCM) router daemon (port 19280) integrated through a provider-router path with two model sets; the OmniRoute gateway (port 20128) translating the native Gemini protocol to an OpenAI-compatible fcm node; a keyless "free" image backend (FLUX.1 Schnell via Hugging Face Space, Pollinations.ai, and Qwen-Image) with layered failover; and a shell-level gemini() automation. Validation showed healthy model probes (785-3,969 ms), FLUX image generation in ~6-8 s (1024^2 WebP), Pollinations in ~3 s, and Qwen-Image in ~10-35 s with verified failover. A security evaluation of a cookie-session "free" proxy concluded it is unsuitable because it violates the Terms of Service, risks credential exfiltration, ships an unidentified binary, and does not solve the official quota constraint.