Logic-Loom v0.1.0-alpha.3
Pre-releaseLogic-Loom v0.1.0-alpha.3
Third alpha — the Research Assistant learns, LLM-backed modes turn on, and
MechDSL joins the app as a Taichi-free FEM code generator.
Builds on v0.1.0-alpha.2 with a full Learn workflow, host-driven LLM
learning modes, and a new MechDSL computational-mechanics bridge — plus a
large zotero-summarizer upgrade.
Re-cut (2026-06-17): binaries rebuilt from
masterafter merging the
vendored-package refresh —akms-learnadvanced to current AKMSmain(the
real-graph-audit lesson fixes) andmechdsl-coreto MechDSL0.2.0. No
app-feature changes versus the original alpha.3 cut.
Platform availability
Multi-platform, built by the GitHub Actions release.yml workflow on real
macOS, Linux, and Windows runners.
- macOS Apple Silicon —
.dmg - Linux x86_64 —
.deband.AppImage - Windows x86_64 —
.msiand NSIS.exeinstaller
Per-platform SHA-256 manifests ship alongside the binaries as SHA256SUMS files.
What's new since v0.1.0-alpha.2
Research Assistant — Learn tab
A new Learn tab compiles AKMS knowledge into learning artifacts (outlines,
anthologies, notebooks, assessments) through the bundled akms-learn engine.
The UI is capability-driven: available modes and exporters are advertised by the
backend, and unavailable or planned modes render disabled instead of failing on
submit. Includes mode-specific forms, Pass B preview panels, artifact history
with an open-export-folder action, a domain-pack status card, a "Send to Learn"
handoff from the Graph view, and a Learning Source Review Bundle export.
LLM-backed learning modes
Modes that need a language model (e.g. expanded/adaptive generation) now turn on
when the host application supplies an LLM provider — the provider is injected
into each learning request, and the mode catalog reflects real host availability
so a provider-less install degrades gracefully.
MechDSL — computational-mechanics bridge
A new MechDSL mode compiles % mechanics LaTeX problem statements into FEM
solver code and transpiles algorithm boxes, with an Overview / Compile /
Artifacts UI. It ships without the Taichi runtime: the emit and transpile
paths are Taichi-free, so the capability is available out of the box and only
the optional on-device verification path would require the extra. This keeps
the bundled engine lightweight.
Vendored dependency upgrades
- zotero-summarizer: large upstream merge adding the NotebookLM pipeline
(zsum nlm) as an LLM backend for card/digest generation, and section-aware
PDF extraction (extraction.mode: sections) using TOC/heading detection. - akms-learn vendored at current AKMS
main— LLM provider seam plus the
real-graph-audit lesson fixes (section heading matching, assessment
references, and section-based pitfalls). - MechDSL (
mechdsl-core+algo2code) vendored at0.2.0to back the
MechDSL bridge.
Desktop
- Light / dark / system theme toggle.
- Translucent Logic-Loom backdrop behind every route.
Verifying downloads
shasum -a 256 -c SHA256SUMS.txt