Tools and Infrastructure
What sits under the products: libraries I build on, tools I run every day, and where compute and model calls come from when the budget is one person’s. How little infrastructure I actually run is in infra-two-tools; my own open-source tools are on /projects/.
Rust
- awesome-rust: the community catalog of Rust libraries
- actix-actor-framework: typed actors with supervision, the base of actix-web
- dano-media-checksums: checksums of the media streams, stable across renames and container swaps
- auto-montage-principles: the five jobs of any auto-montage engine, six principles from building one, and how an agent calls Life2Film’s engine through Studio’s MCP door
- deepwiki-rs: C4 architecture docs generated from a codebase
- rustfs-object-storage: S3-compatible storage, 2.3x faster than MinIO on small objects
- native-frontend-toolchain: the JS toolchain rewritten in Rust and Go, 10 to 100 times faster
Compute and models
- cheap-compute-cpu-gpu: measured prices for burst CPU, a small CUDA job and single-GPU training
- llm-providers-credits: free tiers, startup credits and open models
- cartesia: speech synthesis, transcription and voice agents; the 12-month startup grant and where it goes
- codex-oauth-proxy: LLM calls through a ChatGPT subscription, no API key
- rlhf-book: the free textbook on RLHF, from preference data to DPO and RLVR
Search for agents
- own-searxng-for-agents: your own SearXNG with a Tavily-shaped API as web search for Claude Code, and the engine lessons with numbers
- zero-trust-private-network: a private network on Cloudflare Zero Trust instead of a VPN: tunnels, Access logins for people and agents, SSH without keys, WARP private IPs
The terminal
- tv-terminal-search: a fuzzy finder over this wiki, docs and stacks
Google accounts for agents
- gmail-cli-several-mailboxes-gws: several Gmail mailboxes for a coding agent with the gws CLI, read-only, and the 7-day OAuth Testing trap
- youtube-channels-from-terminal: upload and manage YouTube channels with the Data API, quota math, one token per channel