Field guide · for classroom & independent use
Context Infrastructure Field Guide
Six tools for working on the layer beneath AI systems — what they remember, what they retrieve, and what they're allowed to do. Free to try; feedback welcome.
Each entry below has its own guide: what the tool actually does, what it needs, a step-by-step walkthrough, the exact output you should see, and what to do when it misbehaves. No account is required for any tool marked Ready. Try one, try all six — and if something breaks, confuses you, or seems pointless, that's exactly the feedback we're after.
Know which files matter before you touch the code.
Scores and bundles the most relevant files for a coding task before you touch the code — then tells you how well it predicted what you actually changed. Runs locally; nothing leaves your machine on the free tier.
Catch a bad training set before it costs GPU hours.
A CI safety layer for training-data changes: catches structurally risky datasets and metadata before you spend GPU hours on a bad training run.
Build a context-aware app against realistic mock data.
Build a small context-aware application — meeting prep, an account brief — against realistic mock data, no cloud account required.
A bounded execution envelope for AI coding agents.
A local runtime that gives AI coding agents a bounded, auditable execution envelope — capability manifests, denied-action receipts, no host-level changes.
Compile the minimum sufficient context before the prompt goes out.
A governance SDK for agent runtimes: compiles a minimum sufficient context package before a prompt goes to an LLM, scrubs secrets, and enforces tool boundaries via a local sidecar.
Prompt safety coaching that never leaves your machine.
A local-first companion for AI prompt safety: checks browser AI prompts before they're sent, coaches you when a request looks risky, and keeps a privacy-preserving receipt — all on your own machine.
Questions or bug reports — reply to the inviting email, or open an issue on the linked repo.