Agent-readable docs index: /llms.txt. Full docs in one file: /llms-full.txt. Download /docs.zip to grep all markdown files locally.

For your AI coding agent

KINDI ships a single canonical instructions file, kindi.agents.md, that teaches your coding agent the API surface, the exact crypto parameters, the token formats, the error catalogue, and the anti-patterns agents reliably get wrong. Drop it into your repo and your agent stops guessing.
The file itself is portable. Only the install path differs per tool.
This page is the equivalent of npx shadcn add button. One file, one paste, your agent now knows KINDI.

Pick your tool

Project-scoped (recommended; committed to the repo):
curl -fsSL https://docs.kindi.me/kindi.agents.md >> CLAUDE.md
Or user-scoped (every Claude Code session, all repos):
mkdir -p ~/.claude/skills/kindi
curl -fsSL https://docs.kindi.me/kindi.agents.md \
  > ~/.claude/skills/kindi/SKILL.md

Verify the skill is loaded

Once you've installed it, ask your agent:
Without fetching anything, tell me the exact HKDF salt and info strings KINDI uses to derive the KEK.
A correctly-loaded skill answers masker-kek-salt-v1 and masker-kek-v1. If the agent says anything else, or tries to fetch the docs to answer, the file isn't in its context. Re-check the install path for your tool.

What's in it

The file pins:
  • Surface: base URL, auth header, and the endpoints integrators actually call — POST /api/v1/mask, POST /api/v1/redact, the unauthenticated GET /api/v1/health probe, and the POST /api/v1/proxy/{provider}/… LLM Proxy. It also marks the dashboard-internal routes (/me/*, /v1/keys/*, /auth/*, /admin/*) as ones an agent should leave alone.
  • Decision tree: when to use mask vs redact.
  • JSON contracts: full request + response shape for both endpoints.
  • The crypto dance: exact HKDF salt + info, AES-GCM nonce size, base64 variant, decryption order, with working Python and TypeScript reference snippets.
  • PII type catalogue: every entity KINDI detects, with examples.
  • Error catalogue: every status code, what it means, what to retry.
  • Anti-patterns: the 9 mistakes agents reliably make against KINDI's API (nonce reuse, wrong base64 variant, server-side decryption, mixed token regexes…).
  • A self-check prompt so you can confirm the skill is loaded.

Stay current

The file is versioned in-place by an HTML comment on line 1, of the form <!-- kindi-agents vN · YYYY-MM-DD -->. The current revision is v2. Check the first line of your installed copy against the live file to see whether you're behind.
There is no separate update command: re-run whichever install command you used above and let it overwrite the file. Two caveats when you do: the Claude Code and Codex CLI snippets append (>>), so re-running them adds a second copy rather than replacing the first; delete the old block before re-appending. Every other snippet redirects with > and overwrites cleanly.
The URL https://docs.kindi.me/kindi.agents.md is treated as a stable contract; once you depend on it, we don't rename it.

Read it yourself

The raw file is human-readable too: docs.kindi.me/kindi.agents.md.
If you'd rather your agent pull docs on demand instead of pinning this file, point it at docs.kindi.me/llms.txt, a machine-readable index of every page that links to the raw .md view for each. For bulk-grep-able offline access, fetch docs.kindi.me/docs.zip, which bundles every page as a markdown file in one archive.
Powered byHolocron