New features
Local model supportAllKnower can now route parser and embedding calls to local providers such as Ollama and LM Studio alongside OpenRouter. Use thelocal/ or ollama/ prefix on any task model to point it at your local endpoint, configured via LOCAL_PROVIDER_BASE_URL and LOCAL_PROVIDER_API_KEY. See Configuration for setup.Updates
Cloud fallbacks for local modelsWhen a local model call fails, AllKnower automatically falls back to the configured cloud models for that task. Brain dump, autocomplete, and other parsing features stay reliable even when your local server is offline. Configure fallbacks per task with*_FALLBACK_1, _2, and _3.Cleaner local requestsRequests to local endpoints now strip OpenRouter-specific parameters that local providers don’t understand, improving compatibility with Ollama, LM Studio, and other OpenAI-compatible servers.New features
Public sharing portalThe Portal now ships a dedicated public surface for sharing lore with players and readers. Visitors can browse published entries and open rich detail pages with portraits, theme songs, attributes, and resolved relations—all without signing in. See Sharing lore with players for setup.Public searchReaders on the public portal can now search across published lore from the landing page. Results are scoped to the share tree, so only intentionally published content is discoverable.Public image servingPortraits and other images attached to shared lore now load on public pages through a dedicated image endpoint, so shared entries render with full visuals for unauthenticated visitors.Updates
Shared content browserThe Portal’s shared content view now has clearer Published, Draft, and GM-only filters, inline quick-toggle controls, and a search box. Audit what’s live and flip publish state without leaving the page.Share settings panelThe per-entry share panel now surfaces the computed share URL, a custom alias slug, and optional password protection in a single card, so configuring how a lore entry is published is a one-stop action.Refreshed brandingThe Portal, landing page, and favicon now use the new grimoire logo, replacing the previous generic scroll icon across navigation, sidebar, and browser tabs.Fixes
- Public share visibility — Resolves correctly so published lore is reliably reachable from public URLs.
- Share settings audit — The settings panel shows a clear warning when a note isn’t in the share tree, preventing confusion about why a “published” note isn’t visible.
- Hardened owner auth failure paths — Lookup failures return a controlled
503and bootstrap sign-up now fails cleanly if owner promotion cannot resolve the created user. - Anonymous owner lookups — Short-circuit instead of doing unnecessary work, improving response times for unauthenticated requests.
New features
Public lore pagesPublished lore now has its own public, read-only pages and endpoints that anyone can view without signing in. Share a link and visitors can read the entry directly—no account required.Owner loginThe Portal now has a dedicated owner login page that gates access to private dashboards, settings, and shared views. Unauthenticated visitors are redirected to login for private surfaces and see published lore and the login page only.Publication status in the share treeThe share tree now shows explicit Published and Not Shared indicators, making it clear which entries are live and which are private.Updates
Single-owner PortalWe removed self-service sign-up and registration from the Portal and Settings pages. Login requires an existing AllKnower owner session before authentication cookies are issued. Private APIs return401 for requests without owner credentials.AllKnower owner gateAllKnower enforces a single-owner gate, aligning its authentication model with the Portal.Simpler portal middlewareWe simplified access control for portal routes, simplifying redirects between public lore, login, and private pages.Fixes
- Share links point directly to the destination URL instead of a constructed path, landing visitors on the correct page.
- The system bounds and rounds third-party model pricing from OpenRouter before storage. This resolves an issue that caused pricing updates to fail for some models.
New features
Relationship graphThe Portal relationship view renders with React Flow instead of Mermaid, providing pan-and-zoom, cleaner layouts, and consistent relation colors across the app.Push notificationsSubscribe to push notifications from the Portal to receive alerts when parsing jobs, suggestions, and shared updates finish. A notification bell is now in the top navigation.Streaming AI responsesAI responses stream token-by-token in brain dump and entity views, displaying results during generation instead of waiting for the full response.Auto-relate on brain dumpWhen you save a brain dump, the system proposes relationships to existing entities. Suggestions are cached by content hash and deduplicated in flight, making repeat queries faster.Updates
Hybrid searchSearch blends keyword and vector results using Reciprocal Rank Fusion, surfacing relevant matches for paraphrased or conceptual queries.Relation suggestions cacheThe system caches and invalidates suggestion responses by content hash, with inflight deduplication. This increases suggestion speed and reduces latency.AllCodex Core clarified as server-onlyDocumentation updates clarify that AllCodex Core is a server-only fork (no desktop client), runs on Node 22, and ships as a fixed set of packages.Fixes
- Fixed relation color sync between AllKnower and the Portal so colors stay consistent across views.
- Fixed a streaming entity parsing issue and tightened Zod null coercion to prevent intermittent parse errors.
- Fixed statblock rendering glitches in entity views.
- Scoped Phase 5 graph queries by
userIdto prevent cross-tenant data exposure. - Tightened rate limiter scoping and hardened integration tests.