Changelog

Notable, user-facing milestones. For the complete commit-by-commit history, see the repository itself.


Recent

  • Generating a dashboard config can now check before replacing one that’s already there: every generate path (data API, tenant console, platform admin, and the Config Editor’s own “Regenerate”) used to overwrite an existing file silently, re-running the metered AI call every time even when nothing needed to change. A new ifExists parameter (ask | replace | use_existing) lets a caller check first — the default stays replace, exactly today’s behavior for every existing caller, so overwrite is never blocked, only ever opt-in-protected. A new config-file lookup (data API and developer API) can check without ever generating. See API reference → Database & schema endpoints.
  • A workspace’s own role now applies from a LoginLink sign-in regardless of the “own roles” setting (WebSvc.CrudAdmin 2.1.7): the fix just below (2.1.6) introduced a regression — a role created directly for a tenant by a platform admin (bypassing tenant self-service) stopped being recognized from a LoginLink sign-in for any workspace that hasn’t explicitly turned “own roles” on, the default for a new workspace. That setting governs building a role through tenant self-service; it never governed using one that already exists. Fixed without touching tenant self-service role assignment, which is unaffected.
  • A LoginLink sign-in now gets its built-in role’s permissions applied (WebSvc.CrudAdmin 2.1.6): a person signing in through LoginLink with a built-in role (Reader, Writer, Editor, Database admin) never actually got that role’s permissions — the lookup only ever matched a workspace’s own custom roles, so every real data call through a LoginLink-authenticated app failed with a permissions error. Resolved the same tested way the admin console and key-assignment paths already do; a LoginLink token still never grants Owner. See Features → LoginLink integration.
  • A config generated through the data API is now recorded the same way an admin-generated one is: generating a config via GET .../database/generate-config (the route a tenant’s own app calls, not the admin console’s own button) pushed the file to GitHub correctly but never updated the database’s own row — so config_file_url stayed empty and a caller reading that row couldn’t find the file it had just published. Both paths now record the same way. No caller-side change needed; the field simply starts showing up.
  • Generate Config was broken outright for a short window after the no-restart fix below shipped: a DependencyInjectionException (“No service for type ‘AI.Providers.OpenAIProvider’ has been registered”) on every call. A regression in that same fix — removing the only thing that fed the AI provider type into app configuration at startup also broke a different package’s own internal registration that depended on it. Fixed by building the AI provider directly instead of through that indirection; pinned with a new regression test so this exact failure mode can’t silently return.
  • The Users page shows a person’s email, not just their raw subject id (WebSvc.CrudAdmin 2.1.5): a LoginLink-authenticated person with no display name set used to show only their opaque subject id forever — their email is already resolved on every LoginLink data-API call, it just never reached the record this page reads from. Threaded through the existing usage pipeline instead of a new lookup: an existing person’s row is backfilled the next time they make a request, never overwriting a display name or email already set. See Tenant console → Users.
  • GitHub, AI, and Notify Host Settings no longer require a restart: these three groups used to be read from CrudAdmin’s database exactly once, at boot, into a frozen snapshot — changing a GitHub token, the AI provider/model, or Notify’s base URL silently kept using the old value until the process restarted. They’re now read fresh from Host Settings on every use, the same live behavior every other setting on this host already had. Also fixed in the same pass: leaving the AI URL field blank previously fell back to a built-in default that didn’t match the request CrudApi actually sends, failing every Generate Config call outright — it now defaults to the correct endpoint. The one remaining exception: the separate Config Editor (WebSvc.ConfigVisualizer) still reads its own copy of the GitHub config-path-prefix once at startup. See Admin console → Settings.
  • Auto-approve LoginLink tenant admins into the console (WebSvc.CrudAdmin, pending release): a tenant’s own console previously turned away every new LoginLink sign-in until an already-linked person approved them from Access → People — real friction for someone who’d already proven they’re an admin of the same connected LoginLink organisation. “Who may sign in” is now a 3-way choice: the existing strict (approve everyone) and open (approve no one) modes, plus a new middle option that auto-approves any sign-in whose LoginLink account is a tenant_admin in that organisation. See Features → LoginLink integration.
  • Platform address becomes a Host Setting (WebSvc.CrudAdmin 2.1.3): CrudAdmin’s own public address — behind its JWT issuer/audience and the LoginLink API identifier — no longer has to be set via the Public:BaseUrl environment variable. It’s now a live Host Setting (editable with no restart) also captured during first-run setup; an existing deployment that already sets the env var needs no action, since the setting simply takes precedence when present. See Admin console → Settings.
  • Registered-API identifier and registration-failure reporting fixed (WebSvc.CrudAdmin 2.1.2): the identifier CrudApi registers itself under in LoginLink now reliably matches the one it validates against — it prefers the deployment’s own Public:BaseUrl, rather than falling back to whichever address served an individual request, which could differ behind a reverse proxy and refuse correctly addressed tokens intermittently. A failed “register CrudApi as an API” attempt is also no longer silent: the reason is kept and shown on the Tenant console’s Access page (and in GET /api/tenant/loginlink/data’s apiRegisterError) until a later attempt succeeds. See Features → LoginLink integration.
  • CrudApi as a registered API in LoginLink (WebSvc.CrudAdmin 2.1.0–2.1.1): a tenant’s own separate app (not just the one app CrudAdmin itself registers) can now call the data API as its signed-in users — LoginLink grants it access to CrudApi as an API (a resource server), and it requests a token addressed to CrudApi directly rather than needing to be CrudAdmin’s own trusted app. A workspace can now only ever connect to its own LoginLink organisation — never the platform’s top-level one, and never one another workspace already uses. See Features → LoginLink integration.
  • Admin console rewrite (WebSvc.CrudAdmin 2.0.0): the separate Roles, Policies, Users, Provisioning Keys, Issuance Keys, JWT Sessions, Tenant Sessions and API Clients pages are replaced by a unified Access model; a two-tier Developer API key system (platform- and workspace-scoped) replaces ad hoc key handling; LoginLink adds tenant self-service login links as an alternative to password/JWT login; Managed databases adds policy-gated, script-driven database creation; the old Backup and Migrate-to-MySQL pages are replaced by a single Storage & data page. Breaking: those old page routes are gone — see Admin console for where each capability moved to.
  • Unregister a tenant database (WebSvc.CrudAdmin 2.0.1): a database can now be removed from a workspace’s own registry — CORS origins, row policies, rate limits and cached config state go with it — without the underlying database ever being touched. See Tenant console → Databases.
  • Config Visualizer tenant-scoped sessions (WebSvc.ConfigVisualizer 1.1.0): a signed handoff can now mint a session locked to one tenant and database, not only platform-admin. The scope is checked on every request, not just at login, so one tenant can never reach another’s config editor — not even by editing the URL directly. See Features → Generate Config & dashboards.
  • SqliteAdmin per-file read-only protection (WebSvc.SqliteAdmin 1.2.0): specific database files (by name or full path) can be marked browse/query/export-only, independent of making the whole SQLite browser read-only.
  • Generate Config now reports publishing status honestly — a failed GitHub push no longer looks identical to a full success — and gained an admin-triggered route plus a configurable path. See Features → Generate Config & dashboards.
  • Config Visualizer shipped — a visual editor over a generated dashboard config, including a “Modern” non-technical editing mode. See Features → Generate Config & dashboards.
  • Roles export endpoint added, feeding LoginLink’s role-sync connector and any other integration that needs this platform’s real role names externally.
  • Rate limiting (WebSvc.RateLimitAdmin) wired in as a companion surface, reached by signed handoff from a database’s own Actions menu. See Features → Rate limiting.
  • Runtime data consolidated into a single App_Data folder — previously scattered across separate Data/, RateLimitData/, ConfigVisualizerData/, and bulk-files/ locations, each needing its own Docker volume. See Deployment → Docker.
  • Database-scoped routing: the data API’s address changed from /api/v1/{db_key}/... to /api/v1/{tenant}/{db}/..., since a database key only needs to be unique within its own tenant, not globally. A breaking change to v1 with no parallel v2 — every caller needed the added tenant segment.
  • The first-run setup wizard shipped, gating the entire app behind creating a real first admin account. See Deployment → Setup wizard.
  • Credential-exposure hardening: a committed credentials file was removed from the working tree and .gitignore-blocked going forward; forwarded-header trust became configurable (ReverseProxy__TrustedProxyIPs) instead of unconditionally trusting any proxy.

Earlier

  • Generic Docker Hub/GHCR deployment documented, alongside the original Plesk-specific walkthrough.
  • Persistent-volume requirements for the data directory first documented, after a real deployment lost data from a missing volume mount.
  • Production Swagger UI fixed to fetch its spec from the correct path (/docs, not /swagger).

For specific open limitations rather than historical changes, see Reference → Known limitations.


Copyright © CrudApi. Distributed under an open license.

This site uses Just the Docs, a documentation theme for Jekyll.