Skip to content

Anubis OSS v3.7.0

Choose a tag to compare

View all tags
@uncSoft uncSoft released this 28 May 21:20
· 62 commits to main since this release
v3.7.0
62e7ae6

Anubis OSS 3.7 — Flow Builder

The headline feature this release is the Flow Builder — a Shortcuts-style editor where you author multi-step benchmark recipes (Set Backend → Set Model → Repeat × 5 → Run Benchmark → Unload), save them, share them as .anubisflow JSON, and play them back hands-off. Every individual run still lands in normal Run History, and the whole sequence produces a single share-ready report card.

New: the Flow Builder tab

  • Three-pane editor: a left Palette of 12 step types, a center Step List as a nestable tree, a right Inspector that swaps controls per step type.
  • Drag from the palette into precise positions, or click to append. Reorder existing steps by drag, by ↑/↓ chevrons, or via the inspector.
  • 12 step types: Set Backend / Model / Prompt / Parameters, Run Benchmark, Repeat × N, For Each Model, For Each Prompt, Unload Model, Reset Connection, Cool Down (fixed or thermal), Annotate.
  • Live run count chip in the editor header multiplies through containers — 3 models × 2 prompts × 5 reps = 30 runs.
  • Live lint: dangling Set steps, empty For Each lists, blank entries, missing prerequisites are all caught with a yellow ⚠ on the offending row and a clickable count chip in the header.

For Each, explicitly

  1. For Each replaces the corresponding Set. Inside a For Each Model, the active model is whatever the iterator just picked — any prior Set Model is ignored for that iteration (the lint catches this).
  2. Nested containers multiply. Total runs = outer × inner × deeper. 3 models × 2 prompts × 5 reps = 30 BenchmarkSessions.

Six built-in templates

Sidebar + → New from Template: Quick Smoke Test, Repeated Run (5×), Cold vs Warm Start, Multi-Model Comparison, Q4 vs Q8 Quantization, Prompt Variants. Each is a real working flow you can edit or use as-is.

Run sheet + share-ready report

  • Run sheet shows live per-step progress with checkmarks, a scrollback log, a "Run X of Y" counter, the latest session's stats, and a Stop (⌘.) that cancels mid-run cleanly.
  • Report opens to a forced-dark 1920×1080 card with hero, winner callout, per-model leaderboard with relative bars, per-rep tables, and a methodology footer.
  • Toggle to 1080×1080 for Instagram/X. Save as PNG @ 2× retina (3840×2160 or 2160×2160), copy to clipboard, or save per-model CSV.
  • Run History in the Flows sidebar lists past flow runs; one click reopens the report.

Portable .anubisflow files

Right-click any flow → Export… writes a versioned JSON file. + → Import .anubisflow… reads it back. Older clients refuse newer files instead of silently misreading them.

In-app help

The ? button in the Flows sidebar header opens a long-form help sheet covering the editor, every step type, the two For Each rules with worked code-style examples, templates, lint, and how to read reports.

Polish

  • Editor edits (add / reorder / delete) ease in/out with a 0.22 s animation; drop targets light up early with a generous 14 pt catch area so you don't have to land on a pixel band.
  • Inspector model pickers in Set Model and For Each Model are backend-aware — they filter to the backend implied by the most recent Set Backend step at that point in the flow.
  • "Clear All History" in the Run History menu, with a confirmation that spells out that the underlying BenchmarkSessions survive on the Benchmark tab.

Download

Download Anubis-OSS-3.7.zip, unzip, and drag to /Applications. The app is signed and notarized.

Or:

brew install --cask uncsoft/anubis/anubis-oss

Existing installs will auto-update via Sparkle.

Requirements

  • macOS 15.0+ (Sequoia)
  • Apple Silicon (M1 / M2 / M3 / M4 / M5+)
  • At least one inference backend (Ollama recommended)
Assets 4
Loading