I'm building Bulwark — an open-source LLM API cost-guard proxy.
Hard daily and monthly USD caps. Bedtime Mode that trips when today's spend hits 2× your rolling baseline during sleeping hours. Exact-match caching with semantic-cache scaffolding. Cloudflare Workers. AGPL-3.0.
The goal is simple: stop your AI bill before it stops you.
🛡️ Bulwark — LLM cost-guard proxy (OpenAI, Anthropic, Gemini). v1 shipped, v1.1 in flight (roadmap).
📐 ApexForm — precision CAD / 3D modelling consulting. Reverse engineering, surface modelling, jig design. The day job that funds the proxy.
🎓 Online MBA in flight. Lean Six Sigma in parallel.
The split between agent-runtime budget enforcement and gateway-level cost guards. Both are valid, both have different failure modes, and the honest production answer is probably both with a clean handoff. See the open Discussion on Velocity and Bulwark's DESIGN.md if you want the long version.
LLM cost infrastructure · agent runtimes · API rate limits · circuit breakers · distributed budget state · provider fallback chains · semantic caching · anything in the "AI bills are out of control" space.
Open an issue, start a discussion, or fork the proxy and tell me what's broken.
📍 Halls Head, Western Australia · 🇦🇺