Skip to content

Releases: liana-banyan/mnemosyne

MnemosyneC v0.1.25 — Offline-Ready Pre-Bundle

02 Jun 01:17

Choose a tag to compare

v0.1.25 — Offline-Ready Pre-Bundle (BP070 Scope 1)

What's new

  • Bundled gemma2:2b + ollama.exe inside the installer — zero first-run download required
  • Offline out-of-box: no internet connection needed to start your private AI
  • Installer size: ~1.6 GB (includes 1.6 GB gemma2:2b model + Ollama engine)

Install

Download MnemosyneC-Setup-0.1.25.exe and run. The AI engine starts immediately — no downloads, no setup.

MnemosyneC v0.4.2 — Plow Worker Survival + Hardware-Tier Auto-Select + Atomic Deploy

15 Jun 21:39

Choose a tag to compare

v0.4.2 — Stability Gate BP083 SEG-1/2/3/3.5/4/5 — Worker survival · hardware-tier auto-select · atomic deploy · onboarding lifecycle · Plow resume

v0.1.60 — Shadow E-Giant Concordance Live + Full Plow Loop

13 Jun 06:48

Choose a tag to compare

Real 3-voter Shadow E-Giant concordance (correctness/consistency/coverage · gemma4:12b). Plow the Field 14-domain UI. Substrate Stats Tab. LB Membership scaffold. MCP Bridge scaffold. Browser extension scaffold. R3 Andon integrity. Clipboard/hotkey capture.

v0.1.58 — Ask cold-start heartbeat + token streaming

13 Jun 05:40

Choose a tag to compare

MnemosyneC v0.1.58

Replaces v0.1.57.1 (4-part semver naming fix — auto-update now works correctly for users on v0.1.57).

What's in this build

  • Ask cold-start indicator: 🧠 Gemma is loading into memory (first run, ~30–90s)... (Xs)
  • Ask thinking indicator: Thinking... (Xs) on subsequent inferences
  • Token streaming: tokens render as they arrive via Ollama stream:true
  • AbortController: in-flight stream cancelled on new send
  • Differentiated error messages: preload missing / query threw / Ollama unreachable

v0.1.57 — Staging (smoke-test before Latest)

13 Jun 04:03

Choose a tag to compare

MnemosyneC v0.1.57 — STAGING

STAGING release. Do not install unless directed by the team.

What's in this build

  • SEG-1 Substrate R2 HOT retrieve path — Gemma now reads verified eblets BEFORE answering; LeanAskTab routed through IPC substrate path (was direct Ollama fetch); [SubstrateHOT] telemetry counter
  • SEG-2 Test It Out tab (🧪) — 5-question substrate-warming workout; correct answers write to eblet store; per-question progress + run history
  • SEG-3 Renderer TS cleanup — 41 pre-existing renderer errors → 0 (fixed type declarations, ES2023 compat, PhoebePage conflict)
  • SEG-4 Carry-along — version bump to 0.1.57, Cephas version.json lag fixed (was at 0.1.55)

New VERIFY discipline (BP081 canon)

Binary install verify now required: check diagnostic log shows correct App version before declaring smoke green.

Note

v0.1.55 and v0.1.56 remain at Pre-release. v0.1.57 is a superset — first clean Founder install smoke will promote this to Latest.

v0.1.57.1 — Ask cold-start heartbeat + token streaming

13 Jun 05:26

Choose a tag to compare

Superseded by v0.1.60 — 4-part semver auto-update issue. Do not install.

v0.1.56 — Staging (M0/M1/M2/M3 smoke-test)

13 Jun 02:56

Choose a tag to compare

MnemosyneC v0.1.56 — STAGING

STAGING release for M0/M1/M2/M3 smoke-test. Do not install unless directed.

What's in this build

  • SEG-1 LEAN Ask model-detection family-match — fixes silent failure on machines with pre-existing Gemma variants (gemma2:12b, gemma:12b, etc.); auto-selects best-fit model by family/size/version ranking
  • SEG-2 Gemma 4 12B progressive auto-pull — first-launch pull UI with progress bar, GB/ETA display, 2s heartbeat, cancel button; fires when no compatible model found
  • SEG-3 Connect Via Invite Token Availability tab — renamed button + full Cue Deck Share tab: MY card (copy/QR/mesh-send) + RECEIVED cards (explicit empty state, paste form)
  • SEG-4 Substrate accumulator R1 — Spider/Sprite accept handlers now write verified eblets to local JSONL store; append-only, Andon-gated, sha256 stamped
  • SEG-5 Carry-along sweep — federation:connect-peer IPC, silent email stub, dispute types, CI token confirmed current

Adversarial audit (Sonnet 4.6 VERIFY SEG)

All 4 code-changing SEGs passed adversarial audit. 0 new renderer TS errors. Main TS clean. assert-ollama: ollama.exe + floor model + vc_redist all present in installer.

v0.1.55 — Staging (M0/M1 smoke-test)

13 Jun 01:35

Choose a tag to compare

Pre-release

MnemosyneC v0.1.55 — STAGING

STAGING release for M0/M1 smoke-test. Do not install unless directed.

What's in this build

  • SEG-1 OllamaManager singleton — fixes 'Could not reach local AI' on LEAN Ask tab (all code paths now route through one health-check)
  • SEG-2 NSIS LAN binding — install-time sets OLLAMA_HOST=0.0.0.0:11434 system-wide for LAN federation; HKLM→HKCU fallback; 2 Composer-2.5 NSIS bugs caught and fixed by Sonnet 4.6 adversarial audit
  • SEG-3 COMMUNITY-CONNECT — first-launch button with 4 UX states (idle/connecting/success/fail) + heartbeat; connects to FounderDenken seed peer via relay.lianabanyan.com
  • SEG-4 Dr. MnemosyneC mascot icons — v0.1.53.1 assets confirmed folded, path-aligned across 4 surfaces
  • SEG-5 Cooldown decay Option 2 — effective rejections now decay -1 per 30 clean days

Substrate accumulator audit (BP081 SEG-0)

R1/R2/R3 gaps identified for v0.1.55-v0.1.57 backlog. See BISHOP_DROPZONE/BP081_SUBSTRATE_ACCUMULATOR_AUDIT.md.

MnemosyneC v0.1.24 — BP067 Family-Install Release

31 May 22:05

Choose a tag to compare

MnemosyneC v0.1.24 — BP067 Family-Install Release

What's New

  • One-spine first-run flow: Cover → Value → Ask-answer (instant payoff) → Optional folder → Main app
  • Private AI floor: Gemma 2B (Google, non-Meta) auto-installed on first use; zero accounts, zero API keys
  • Transparent install screen: Shows exactly what's being set up (Ollama attribution accurate — Y-Combinator startup, MIT license)
  • Smart Ollama detection: Uses existing Ollama install if present; skips download
  • SaltFighter cover screen: With skip / don't-show-again; auto-advances at 8s
  • Techie surfaces hidden: Gauntlet, mode chip, Bridge banner hidden on first launch (discoverable via 'For Techies' door)
  • No dead-ends: Every screen has a clear exit; folder ask deferred until after first answer

AI Engine

  • Default floor model: gemma2:2b (Google Gemma 2 · 2B parameters · ~1.6 GB · pulled once on first use)
  • Model picker available: Llama, Mistral, Qwen, Phi — model-agnostic via court-router abstraction
  • Upgrade path: Good (~2 GB) or Great (~5 GB) offered only AFTER first answer

Install

  • Download MnemosyneC-Setup-0.1.24.exe and run it
  • On first launch: private AI sets up automatically (requires internet for first-time model download ~1.6 GB)
  • All data stays on your computer — private by default

SmartScreen Note

Windows may show a SmartScreen warning. Click 'More info' → 'Run anyway'. This is expected for unsigned installers.

Free to use. Better to join. Share and Save.

MnemosyneC v0.1.23

31 May 04:26

Choose a tag to compare

MnemosyneC v0.1.23 Phase 1

1A: FirstStepsView - /yr membership, Stripe LIVE, auto-renew
1B: SaltFighterFirstRun - cover splash, HEOHO, folder picker
1C: Atlas week view + Recipes (3 seeds) + Schedule this meal cross-tab link
1D: electron-updater safe tier (autoDownload:false, banner notification)

SHA-256: 125DB0BD63BA4EA589A538DF8411899A413044F6661A6343DD102346B2BC749A
Commit: fcbbe56