Sirius 0.6.2 Beta 2

Sirius 0.6.2 Beta 2

This hotfix repairs the verified updater's restart handoff. Beta 1 could finish
its download and SHA-256 verification but remain indefinitely on RESTARTING
when runtime shutdown did not return.

What changed

• App shutdown now has a five-second graceful deadline. Sirius still asks every
  owned runtime to stop and saves recoverable state, but an unresponsive runtime
  can no longer block application termination forever.
• The external verified installer waits for graceful shutdown, then escalates
  through TERM and KILL before touching /Applications/Sirius.app.
• A failed helper run leaves a persistent, user-visible error that is consumed
  on the next update check instead of leaving the update panel stuck.
• Previous installed bundles are still preserved as timestamped rollback copies.
• The mounted and installed replacement must still pass deep code-signature and
  Gatekeeper verification before Sirius is relaunched.
• Includes all stability, capability routing, bounded-context, Smart Inbox, and
  recoverability improvements from 0.6.2 Beta 1.

Verification

• Updater shutdown-escalation regression test
• Previous-helper-error recovery regression test
• Full Swift test suite
• Optimized release build
• Developer ID signature verification
• Apple notarization and ticket stapling
• Gatekeeper assessment
• SHA-256 release checksum

Package

• Version: 0.6.2 Beta 2
• Build: 30
• File: Sirius-0.6.2-beta.2-arm64.dmg

Sirius 0.6.2 Beta 1

Sirius 0.6.2 Beta 1

This release concentrates on predictable local work: Sirius keeps the fast
Direct Qwen path for ordinary requests, gives fx only the capabilities a turn
needs, and prevents local inference clients from competing for the same model.

What changed

• Direct Qwen, fx, BrowserCode, and model prewarm now share one serialized
  inference lane. This prevents overlapping prefill, duplicate model pressure,
  and another native inference request is already active failures.
• Compiled workspace context now follows the served model's actual context
  window and reserves space for output and tool results.
• Long fx conversations are bounded deterministically while preserving the
  newest complete user and tool turn.
• Capability-based fx selection advertises only the workspace, terminal,
  browser, app-control, specialist, or extended tools required by the request.
• Direct Qwen retains bounded native workspace tools for common file work, so
  fx is not started when the model can safely complete the task itself.
• Recoverability contracts and an inspectable log now record model swaps,
  memory-scope changes, context truncation, and interrupted specialist handoffs.
• KITKAT cross-project context remains read-only while reviewed proposals can
  be written only to the governed Sirius Memory channel.
• The fx diagnostic log is privacy-safe, line-bounded, and rotates at 1 MiB.
• Smart Inbox requests remain isolated, hash-checked, and review-only.

Verification

• Full Swift test suite
• Optimized release build
• Developer ID signature verification
• Apple notarization and ticket stapling
• Gatekeeper assessment
• SHA-256 release checksum

Package

• Version: 0.6.2 Beta 1
• Build: 29
• File: Sirius-0.6.2-beta.1-arm64.dmg

Sirius 0.6.1 Beta 1

Sirius 0.6.1 Beta 1

This is a debug-oriented follow-up to 0.6.0 Beta 2. It makes the production
provider boundary explicit so MLX Serve can be tested independently from the
experimental Native MLX and Elastic paths.

What changed

• MLX Serve is the default inference backend for Direct Qwen, fx, and BrowserCode.
• Native MLX is available only with the explicit development opt-in
  SIRIUS_NATIVE_INFERENCE=1.
• The status strip reports the selected backend (MLX SERVE, NATIVE MLX, or
  ELASTIC) instead of presenting ECO as if it were an engine choice.
• Model residency remains lazy so opening Sirius does not load a large model.
• Release metadata, update detection, and the public landing-page download now
  target this 0.6.1 Beta 1 artifact.
• Verified updates can now be installed with a single RESTART & INSTALL
  action: Sirius hands the SHA-256-checked DMG to a short-lived helper, exits,
  preserves a rollback copy, verifies the replacement, and relaunches.

Debug focus

When testing this build, record the backend label, selected model, available
unified memory, loopback health on port 11234, and direct-model tokens/second
before evaluating fx latency. The Flash-Next 4-bit preview remains an explicit
96 GB-class model selection and can fail its memory preflight on a busy Mac.

Package

• Version: 0.6.1 Beta 1
• Build: 28
• File: Sirius-0.6.1-beta.1-arm64.dmg

Sirius 0.6 Beta 2

Sirius 0.6.0 Beta 2

Beta 2 promotes the verified Sirius 0.6 build used with Qwen3.8 Flash Next through MLX Serve on Apple Silicon.

What changed

• Reliable switching between native MLX and MLX Serve model profiles.
• Transactional rollback when a requested model cannot start.
• Stronger download completeness checks for large model repositories.
• Project constellations with compact neighboring stars and private or shared memory scope.
• Streaming token-usage requests so direct model replies display measured decode speed.
• Clearer local-runtime and memory-pressure diagnostics.
• Graceful handoff between recorded inference backends.

Verified working path

• Qwen3.8 Flash Next MLX Serve 4-bit loaded through Sirius.
• Direct warm replies measured up to 77 tokens per second on the development Mac Studio.
• fx workspace execution and optional macOS Control remain explicit opt-ins.
• Model switching, local chat, project stars, and per-response speed display were exercised in the installed app.

Performance varies with hardware, prompt length, model state, tool use, and memory pressure. Agent-task timing is not the same as raw model decode speed.

Requirements

• Apple Silicon Mac
• macOS 14 or newer
• Adequate storage and unified memory for the selected local model

Install

The beta is ad-hoc signed and is not notarized. Open the DMG, drag Sirius to Applications, then right-click Sirius and choose Open on first launch.

Package

• Version: 0.6.0 Beta 2
• Build: 27
• File: Sirius-0.6.0-beta.2-arm64.dmg
• SHA-256: 995b01ceb8cbd054e0b8259b0ca9d0d6249ec62451242ce76ead41e3b453ea06

Sirius 0.6 Beta 1

Sirius 0.6.0 Beta 1

Sirius is a private local AI workspace for Apple Silicon. Version 0.6 turns each project into a constellation of focused threads and connects direct local inference with explicit, reviewable tools.

Main features

• Fast local Qwen 3.8 inference through MLX Serve.
• Project constellations with fresh neighboring threads.
• Private-star or shared-constellation memory scope.
• Bounded KITKAT and read-only ChatGPT archive retrieval.
• Optional fx Agent for workspace tools and longer execution.
• Optional macOS Control and isolated Browser mode.
• Creative handoffs for Suno prompts, LTX video prompts, and ALi render sequences.
• Reviewed PXL Finance draft handoffs.
• Local model setup, status, and selectable runtime profiles.

Connected applications

• MLX Serve: owns fast local model inference.
• KITKAT: provides reviewed, durable project memory.
• ALi Imagine: owns visual generation queues and outputs.
• Suno: receives copy-ready music prompts and lyrics; Sirius does not submit generations.
• PXL Finance: owns financial records; Sirius creates reviewed drafts only.
• BrowserCode: provides isolated, explicitly enabled web research.

Requirements

• Apple Silicon Mac
• macOS 14 or newer
• Adequate storage and unified memory for the selected local model

Install

The beta is ad-hoc signed and is not notarized. Open the DMG, drag Sirius to Applications, then right-click Sirius and choose Open on first launch.

Package

• Version: 0.6.0 Beta 1
• Build: 26
• File: Sirius-0.6.0-beta.1-arm64.dmg
• SHA-256: 4a55f98e8d59001f47b01141b695ea8ee014c7f2e7f430bf54f5105912528bcf

Sirius 0.5.3 Beta 5

Sirius 0.5.3 Beta 5 makes the local-first path faster and adds two substantial creative-memory workflows.

Highlights:

• Bounded read-only retrieval from a selected ChatGPT archive, with chatgpt:// citations.
• Sirius-owned local memory can be written automatically without direct access to the KITKAT vault.
• A structured Suno v5.5 writer expands an idea into a track title, lyrics, section metatags, positive style direction, and negative prompts.
• Faster local-only startup when Mac control, Web, and fx are disabled.
• fx Agent, BrowserCode Web, and macOS Control remain separate, explicit switches.
• Quieter activity reporting and improved Markdown formatting.

Requirements: macOS 14 or later on Apple Silicon.

This beta is ad-hoc signed and is not yet notarized. On first launch, right-click Sirius and choose Open.

SHA-256: 24e69588cbfb3918fdcbe60fe30209eea60b079c2c32b5de33ae43d1a8e128d7

Sirius 0.5.0 Beta 1

First public beta of Sirius for Apple Silicon.

Highlights:

• Local Qwen inference through Sirius-owned MLX runtime
• fx Agent toggle for focused workspace execution
• macOS Control remains a separate, explicit opt-in
• Persistent project workspaces and quieter activity reporting

Requirements: macOS 14 or later on Apple Silicon.

This beta is ad-hoc signed and is not yet notarized. On first launch, right-click Sirius and choose Open. Sirius will guide you through runtime and model setup.

SHA-256: 5f5a3293aa79947201e0eff1994f14ff710df5677c795c3c74c11e9b96b85a00