Repository navigation
v1.0.0
token-forge 1.0.0 — first public release
Private vibe-coding LLM in your own AWS account. Serve open-weight models on 100% spot instances, connect Claude Code with one command, and let placement-score intelligence pick the region for you.
Highlights
tkfCLI —up/down [--purge]/status/model list/seed/placement/config/connect claude/rotate-key(#11, #19, #21, #22)- R10 intelligent placement — ranks candidate regions by 48h placement-score trend → RTT → price (with quota checks), then races capacity acquisition across the top-K regions in parallel, keeping the first to acquire (First-Acquired-Wins) (#19)
- R11 transport security — CloudFront TLS front door (no domain required), ALB origin-verification gate (403 for bypass attempts), optional source-IP allowlist (
-c allowedCidrs=), and API key rotation (#22) - Cost guards everywhere — idle auto-stop, desired=0 restoration on every failure path, CPU-spot weight pre-seeding (measured: 31GB in 8 min for $0.03) (#3, #5)
- Verified model catalog — Qwen3-Coder-30B (recommended), GLM-4.6, Solar-Open2-250B, all validated in production serving; EP flag fix for block-quantized MoE + TP (#6)
- Anthropic
/v1/messagesnative (validated with Claude Code: tools, streaming, thinking) + OpenAI/v1/chat/completions, prefix caching on by default - Docs — English/Korean/Japanese README, architecture guide, PR/FAQ, value proposition with measured numbers (#7, #9, #15, #23, #30)
Install
npm install -g @roboco/token-forge # provides the tkf commandFull Changelog: https://github.com/roboco-io/token-forge/commits/v1.0.0