0.4.3 — try MindBase with one command: npx mindbase-app

The web UI gets its first real distribution channel, plus a website with live demos.

npx mindbase-app

npx mindbase-app

Starts the local server + web UI, opens your browser, and walks you through installing a free local model that fits your RAM. 2.8MB download, Node 20+, --port / --no-open flags. No clone, no pnpm, no API key.

Website

https://frankchu91.github.io/mindbase-llm-wiki/ — how the LLM-wiki pattern works, live demo clips of the approval flow / wiki-status chip / contradiction lint, and all install paths in one place.

Muse Glimmer support

Meta's open-weight 30B is now the best-tier recommendation on 32GB+ Apple Silicon (Ollama ≥0.32.7, MLX). Quick model switcher on the chat composer badge — run qwen3:14b for interactive work and Glimmer for background synthesis. Thinking-model reasoning now streams live in chat instead of looking frozen.

Fixed

• Projects created from the web UI were dead on arrival — they scaffolded the legacy layout and every operation returned 409. Fresh projects now get the full v2 scaffold.
• Long note titles wrap instead of clipping.
• think:false silently discarded Muse Glimmer's reasoning tokens (stream went quiet for 30s+); thinking models now stream their reasoning as progress.
• Small-talk questions in chat no longer get force-fed through wiki citations.

Full details in CHANGELOG.md.

v0.4.0: chore(release): 0.4.0 — full note authoring + wiki-status chip

0.4.0 — write full notes, watch the wiki absorb them

Three releases in three days, combined story:

0.4.0 — Write full notes in the UI

Cmd+N opens a real WYSIWYG editor: headings, GFM, markdown links, [[wikilink]] autocomplete, outline, backlinks
Notes live in your layer (sources/contributors//notes/) — the AI reads them, never rewrites them
Wiki-status chip on every note: ✨ Add to wiki when the wiki hasn't absorbed it yet → Adding… → ✓ In wiki · N pages. Edit the note and it lights up again
Fixed: WYSIWYG edits were silently copying user notes into the AI-owned research layer
0.3.0 — The Web UI runs the core operations

/contribute, /build, /lint, /research in the browser, orchestrated server-side with your configured LLM
Contribute shows takeaways + a checkbox plan — only what you approve gets written
Health view: contradictions, orphans, stale claims, gaps as actionable cards
0.2.0 — Free local models, no API key

Setup wizard detects your hardware, recommends the best Ollama model for your RAM, installs + verifies it
Runs end-to-end with zero subscriptions, zero data leaving your machine
Full details in CHANGELOG.md.