Get ledric running in two minutes with one Node process and one SQLite file.

The mental model: types, entries, ids/slugs/refs, versions, structural vs inline references, assets, locales, environments, and the wire format.

End-to-end walkthrough: ask Claude to set up a schema, seed content, build an Astro site, wire up inline editing.

Same end-to-end walkthrough as Build with an agent, but with a vanilla PHP consumer using Ledric\\LedricClient.

<script> install, data-ledric-ref / data-ledric-field attributes, refAttrs() helpers, auth, behaviour.

Example prompts: project setup, drafting, schema evolution, bulk ops, refactoring, localization.

Per-type locales, localized:true fields, _locale sidecar, fallback chains, locale-specific slugs.

Honest comparison vs Contentful, Sanity, Payload, Strapi, Directus — and when ledric is the wrong choice.

The questions that come up most: production-readiness, SQLite limits, hosting, the agent angle, migration, backups.

What's stable, in progress, planned, and explicitly out of scope. No timelines — just clear lines.

Field types catalogue, defineType() examples, options, validation, schema evolution.

The 20-tool surface for agents — reads, writes, schema, assets, tags. Same shape as POST /rpc.

Two modes: local (--http-mcp, share one daemon across local clients) and public (--public-mcp, OAuth provider for claude.ai custom connectors).

REST routes for reads, multipart uploads, generic POST /rpc, image transforms, slug redirects, error codes.

@ledric/sdk (TypeScript) and Ledric\\LedricClient (PHP, Composer ledric/sdk) — same wire format.

id / ref_key split, db vs local backends, uploads, image transforms, in-place bytes replacement, transforms cache.

Reader and admin keys, key minting, header formats, closed-reads mode, rotation.

What's running when ledric boots: the packages, the storage adapters, the asset pipeline, the inline editor, process lifecycle.

Production shape: CDN in front of /assets, reverse proxy + TLS, env-supplied keys, backups, Postgres / MySQL.