andreimerfu/pllm

pllm is a 17-star, 240-day-old project with zero recent velocity and only 1 fork. The README describes a straightforward LLM gateway—a pattern that is well-established in the industry (Kong, Tyk, AWS API Gateway, Anthropic's multi-provider routing). Go is the right language for this, but the architecture is commodity: request validation, routing rules, multi-provider load balancing, and OpenAI spec compatibility are all solved problems with no novel algorithmic contribution or defensible moat. DEFENSIBILITY ASSESSMENT: Score 3 reflects a working prototype with minimal adoption. The project lacks the network effects, switching costs, or ecosystem lock-in that would make it defensible. It's a clean implementation of standard patterns but trivially reproducible by anyone with Go and HTTP experience. No data gravity, no proprietary models, no irreplaceable community. FRONTIER RISK: HIGH. Frontier labs (OpenAI, Anthropic, Google, Anthropic) are already shipping multi-provider routing and gateway capabilities as part of their platform strategy (e.g., OpenAI's own proxy patterns, Anthropic's API routing). This is not a niche they avoid—it's a commodity feature they actively develop. A frontier lab could build this in 2–4 weeks as an internal service or add it as a managed offering. There is no defensibility advantage that would cause them to integrate pllm rather than build it themselves. IMPLEMENTATION DEPTH: Prototype. Low velocity (0.0/hr), minimal stars, and no evidence of production hardening, observability, fault tolerance, or multi-tenant isolation. The README promises 'enterprise features' but the code suggests an early-stage POC. NOVELTY: Reimplementation. This applies known gateway patterns (Stripe, AWS, Kong) to the LLM API space. That's a valid product idea but not a technical breakthrough. The OpenAI compatibility layer is a direct copy of the spec, not an original contribution.