hwLedger v0.1.0-alpha

The first alpha release of hwLedger: capacity planner + fleet ledger with enterprise-grade architecture.

Release Date: April 2026
GitHub Release: KooshaPari/hwLedger/releases/tag/v0.1.0-alpha

Download

macOS release build pipeline with code signing and notarization

Keyframes (0 — VLM-friendly)

Loading manifest…

Download MP4 Download GIF Keyframes (repo)

macOS (Signed & Notarized)

Intel/Apple Silicon: hwledger-v0.1.0-alpha-universal.dmg (signed, notarized)
Apple Silicon only: hwledger-v0.1.0-alpha-arm64.dmg

Verification:

bash

# Check code signature
codesign -dv --verbose=4 /Applications/hwLedger.app

# Verify notarization staple
xcrun stapler validate /Applications/hwLedger.app

Sparkle Auto-Updates: The app includes Sparkle appcast for automatic updates.

xml

<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0" xmlns:sparkle="http://www.andymatuschak.org/xml-namespaces/sparkle">
  <channel>
    <title>hwLedger</title>
    <link>https://hwledger.kooshapari.dev</link>
    <description>hwLedger Capacity Planner + Fleet Ledger</description>
    <item>
      <title>v0.1.0-alpha</title>
      <pubDate>Fri, 19 Apr 2026 00:00:00 GMT</pubDate>
      <sparkle:version>0.1.0-alpha</sparkle:version>
      <sparkle:shortVersionString>v0.1.0-alpha</sparkle:shortVersionString>
      <sparkle:minimumSystemVersion>12.0</sparkle:minimumSystemVersion>
      <enclosure url="https://github.com/KooshaPari/hwLedger/releases/download/v0.1.0-alpha/hwledger-v0.1.0-alpha-universal.dmg"
                 sparkle:version="0.1.0-alpha"
                 sparkle:shortVersionString="v0.1.0-alpha"
                 length="28541952"
                 type="application/octet-stream"
                 sparkle:dsaSignature="MCwCFFrJhVSiQzzfpeLPB1/5JZa3lElyAhRrr3U0tJCj7DCn7L1wfVb06mR70g==" />
      <description><![CDATA[
        <p><strong>v0.1.0-alpha Release Notes:</strong></p>
        <ul>
          <li>Math core with architecture dispatch (MHA, GQA, MQA, MLA, SSM, sliding window)</li>
          <li>GPU telemetry (NVIDIA NVML, AMD rocm-smi, Apple Silicon macmon)</li>
          <li>HF Hub and GGUF model ingestion</li>
          <li>Fleet ledger with event sourcing (alpha)</li>
          <li>macOS SwiftUI app (alpha)</li>
          <li>CLI planner, probe, ingest, fleet subcommands</li>
        </ul>
      ]]></description>
    </item>
  </channel>
</rss>

Windows (WinUI 3)

x64: hwledger-v0.1.0-alpha-x64.msix (Windows 10+)
ARM64: Coming in next release

Linux (Qt 6)

AppImage: hwledger-v0.1.0-alpha-x86_64.AppImage
Snap (coming soon)
Ubuntu/Debian PPA (coming soon)

Features

Phase 0 Foundation

Rust monorepo with shared crates (phenotype-event-sourcing, phenotype-cache-adapter, phenotype-config-core)
FFI bindings (UniFFI for macOS, csbindgen for Windows, cxx-qt for Linux)
Traceability harness with spec → test → code mapping
Comprehensive test suite (372 unit + integration tests)

Phase 1: Math Core

Architecture dispatch: MHA, GQA, MQA, MLA, Sliding Window, SSM/Mamba, Hybrid, Sink
KV cache formulas derived per attention type
Resident vs. active parameters for MoE models
Live breakdown: Per-layer VRAM and throughput estimates
Reconciliation: Predictions vs. live telemetry

Supports:

Llama (all variants: 7B, 13B, 70B, Code, Chat, etc.)
Mistral (7B, 8x7B, 8x22B)
Qwen (1.5B–72B, all variants including Qwen2.5, Qwen1.5)
Mixtral (all MoE variants)
DeepSeek (7B, 13B, 33B, 67B, V3)
Phi (2, 3, etc.)
GPT-3.5, GPT-4 (config only, no weights)

Phase 2: Ingestion + Telemetry

HF Hub integration: Pure Rust via hf-hub (auth-free for public models)
GGUF support: Quantized weights inspection
Ollama/LM Studio: REST API integration for live model introspection
GPU telemetry: NVIDIA (NVML), AMD (rocm-smi), Apple Silicon (macmon)
Probe service: JSON output with live resource monitoring
Cache adapter: Two-tier LRU + DashMap with TTL

Phase 3: macOS GUI (Alpha)

SwiftUI Planner UI
Live KV cache visualization
Model selector with HF Hub search
VRAM/throughput breakdown per layer
Telemetry dashboard (alpha)

Phase 4: CLI

hwledger plan <model-config.json> — interactive planner
hwledger probe list --json — enumerate devices
hwledger probe watch — streaming telemetry
hwledger ingest hf://model-id — fetch HF Hub config
hwledger ingest ollama://model:tag — Ollama introspection
hwledger fleet register — device provisioning (alpha)
hwledger fleet audit — fleet health check (alpha)

Phase 5: Fleet Ledger (Alpha)

Event-sourced audit log (SQLite, append-only)
Three dispatch modes: Agent-based (mTLS), Agentless (SSH), Cloud APIs
Cost model with spot pricing integration (Vast.ai, RunPod, Lambda)
Device provisioning and health monitoring

Known Limitations

Windows & Linux GUIs: Not yet built. Use CLI on these platforms.
Inference runtime: Deferred to Phase 4. Use vLLM, TGI, Ollama, or MLX.
GUI journeys: XCUITest automation is a placeholder (WP-25, Phase 3+). See GUI Journeys Status.
Web API: REST endpoints are incomplete; fleet dispatch is alpha.
Spot price caching: Cloud provider pricing is queried on-demand; no predictive model yet.

Breaking Changes

None (first release).

Contributors

Author: Koosha Pari (@KooshaPari)
Math core derivations: Based on Phenotype research (see KV Cache docs)
Testing: 372 unit and integration tests across 11 crates

Support

License

Apache-2.0. See LICENSE.

Next Release: v0.2.0 (June 2026) — Windows GUI, Linux AppImage, fleet dispatch improvements.

hwLedger v0.1.0-alpha ​

Download ​

macOS (Signed & Notarized) ​

Windows (WinUI 3) ​

Linux (Qt 6) ​

Features ​

Phase 0 Foundation ​

Phase 1: Math Core ​

Phase 2: Ingestion + Telemetry ​

Phase 3: macOS GUI (Alpha) ​

Phase 4: CLI ​

Phase 5: Fleet Ledger (Alpha) ​

Known Limitations ​

Breaking Changes ​

Contributors ​

Support ​

License ​