v2.8.0 — Import Profiles Preview Validation Design

v2.8.0

This is a docs-only release.

Summary

v2.8.0 polishes the future Import Profiles design.

It defines the Import Preview sheet, validation result model, severity levels, error and warning messages, unsupported schema handling, conflict handling, Advanced Launch Options validation, and side-effect boundaries.

No app code was changed in this release.

Added

• Added v2.8.0 Import Profiles Preview / Validation design notes
• Expanded import preview requirements
• Defined validation severity: error, warning, info
• Clarified unsupported schemaVersion handling
• Expanded conflict handling rules
• Clarified Advanced Launch Options validation
• Clarified privacy and local path boundaries
• Updated README and task notes

Import Preview Design

The future Import Preview should show:

• source file name
• schemaVersion
• app
• exportedAt
• total profiles
• valid profiles
• invalid profiles
• warnings
• per-profile validation status
• conflict status
• planned action

Validation Boundary

Future import validation should reject or warn on:

• invalid JSON
• unsupported schema versions
• missing profiles array
• empty profile names
• empty model IDs
• invalid hosts
• invalid ports
• invalid probability values
• invalid chatTemplateArgs JSON
• duplicate names
• secret-looking fields
• executable path fields
• local path-like fields

Safety Boundary

Import Preview must be side-effect-free.

It must not:

• start servers
• stop servers
• restart servers
• call /v1/models
• make external HTTP requests
• change selected profile
• change adopted external server state
• download models
• delete files
• change process ownership

Current Implementation Status

Implemented:

• Export Profiles

Not implemented:

• Import Profiles
• Import Preview
• profile validation UI
• conflict handling UI
• skip / rename / replace actions

Direct Mode Boundary

Direct Mode remains unchanged:

OpenAI-compatible client -> mlx_lm.server or adopted external server -> MLX model

MLX Server Manager does not proxy inference requests, act as a chat UI, download models, delete models, or become a multi-backend wrapper.

Current Binary Asset

The current downloadable app binary remains the v2.7.0 unsigned build:

MLXServerManager-v2.7.0-unsigned.zip

This v2.8.0 release does not include a new binary asset.

v2.7.0 — Export Profiles

v2.7.0

This is an app release.

Summary

v2.7.0 adds Export Profiles for model profile metadata.

Users can now export their MLX Server Manager model profiles to a JSON file for backup or migration preparation.

This release implements export only. Import Profiles remains future work.

Added

• Added Export Profiles... action in the Models sidebar
• Added JSON export for model profile metadata
• Added export privacy summary
• Added export success / cancel / failure messages
• Added export logging
• Added ModelProfileExportService
• Updated README and docs for Export Profiles status

Exported JSON

The exported JSON includes:

• schemaVersion
• app
• exportedAt
• profiles
• profile name
• profile modelID
• profile host
• profile port
• non-empty Advanced Launch Options

Not Exported

Exported profiles do not include:

• API keys
• tokens
• secrets
• executable paths
• local model paths
• model weights
• model caches
• logs
• PID
• memory metrics
• readiness results
• runtime process state
• selected current target state
• adopted external server state

Safety Boundary

Export Profiles is side-effect-free.

It does not:

• start servers
• stop servers
• restart servers
• adopt external servers
• forget external servers
• call /v1/models
• make external HTTP requests
• change process ownership
• proxy inference requests

Direct Mode Boundary

Direct Mode remains unchanged:

OpenAI-compatible client -> mlx_lm.server or adopted external server -> MLX model

MLX Server Manager does not proxy inference requests, act as a chat UI, download models, delete models, or become a multi-backend wrapper.

Import Status

Import Profiles is not implemented in this release.

Future import work should include:

• import preview
• schema validation
• profile validation
• conflict handling
• skip / rename / replace options
• no automatic server start

Build

Unsigned macOS app build:

MLXServerManager-v2.7.0-unsigned.zip

SHA-256:

8877aa7912585e733b9b26cc642a3a7a8e09abddae0ca402f0a7d868cef777f6

v2.6.0 — Model Profile Import Export Design

v2.6.0

This is a docs-only release.

Summary

v2.6.0 adds the Model Profile Import / Export design.

The goal is to make future model profile backup, migration, and sharing safer by defining the file format, validation behavior, conflict handling, and security boundaries before implementation.

No app code was changed in this release.

Added

• Added docs/model_profile_import_export.md
• Added README links to the new design document
• Updated onboarding guidance notes
• Updated task notes for v2.6.0

Design Scope

The design covers:

• JSON export format
• versioned schema
• profile metadata export
• import preview
• schema validation
• profile validation
• conflict handling
• Advanced Launch Options handling
• local path handling
• UI flow
• error messages
• testing plan
• future work

Safety Boundary

The design keeps import/export limited to model profile metadata.

It does not include:

• model weights
• model caches
• API keys
• Hugging Face tokens
• GitHub tokens
• secrets
• executable paths by default
• automatic server start
• external process ownership changes

Direct Mode Boundary

Direct Mode remains unchanged:

OpenAI-compatible client -> mlx_lm.server or adopted external server -> MLX model

Model profile import/export is a metadata management feature. It does not proxy inference requests, add a chat UI, download models, delete models, or become a multi-backend router.

Future Implementation Notes

A future implementation should include:

• Export Profiles action
• Import Profiles action
• Import Preview sheet
• conflict handling options: skip, rename, replace
• privacy warning before export
• validation before import
• no automatic process launch after import

Not Changed

• No Swift changes
• No Xcode project changes
• No build setting changes
• No new app binary
• No new zip asset
• No inference proxy
• No chat UI
• No model downloader
• No model deleter
• No multi-backend wrapper

Current Binary Asset

The current downloadable app binary remains the v2.4.0 unsigned build:

MLXServerManager-v2.4.0-unsigned.zip

This v2.6.0 release does not include a new binary asset.

v2.5.0 — Onboarding Guidance Screenshot

v2.5.0

This is a docs/public polish release.

Summary

v2.5.0 adds a README screenshot for the First-run Onboarding Guidance panel introduced in v2.4.0.

No app code was changed in this release.

Added

• Added screenshots/onboarding-guidance-v2.5.png
• Added First-run Onboarding Guidance screenshot to README
• Updated screenshot refresh notes
• Updated onboarding guidance notes
• Updated task notes for v2.5.0

Screenshot Coverage

The new screenshot shows the lightweight onboarding guidance panel in the Not Running / Not Connected state.

It helps explain:

• setup hints for first-time users
• managed server startup guidance
• connection settings flow
• Direct Mode reminder
• no inference proxy behavior

Direct Mode Boundary

Direct Mode remains unchanged:

OpenAI-compatible client -> mlx_lm.server or adopted external server -> MLX model

MLX Server Manager does not proxy inference requests, act as a chat UI, or become a multi-backend wrapper.

Safety Boundary

The onboarding guidance panel is informational only.

It does not:

• install dependencies
• download models
• start external processes
• stop external processes
• restart external processes
• proxy inference
• change process ownership

Not Changed

• No Swift changes
• No Xcode project changes
• No build setting changes
• No new app binary
• No new zip asset
• No inference proxy
• No chat UI
• No multi-backend wrapper

Current Binary Asset

The current downloadable app binary remains the v2.4.0 unsigned build:

MLXServerManager-v2.4.0-unsigned.zip

This v2.5.0 release does not include a new binary asset.

v2.4.0 — First-run Onboarding Guidance

v2.4.0

Summary

v2.4.0 adds the initial first-run onboarding guidance panel.

The goal is to help new users understand what to configure next without adding a large wizard, automatic installer, model downloader, chat UI, or inference proxy.

Added

• Added lightweight Onboarding Guidance panel
• Added OnboardingGuidance model
• Added OnboardingGuidanceView
• Added state-specific first-run guidance
• Added Direct Mode note inside the guidance panel
• Updated README and onboarding docs
• Updated task notes for v2.4.0

Guidance States

The onboarding panel provides short guidance for:

• missing mlx_lm.server executable path
• missing or incomplete model profile
• Not Running / Not Connected state
• External Server Detected state
• Adopted External Server state
• Managed Server Running state

Behavior

The guidance panel is informational only.

It does not:

• automatically configure paths
• install dependencies
• download models
• start external processes
• stop external processes
• restart external processes
• change Advanced Launch Options
• change lifecycle safety rules

Direct Mode Boundary

Direct Mode remains unchanged:

OpenAI-compatible client -> mlx_lm.server or adopted external server -> MLX model

MLX Server Manager does not proxy inference requests, act as a chat UI, or become a multi-backend wrapper.

Ownership Boundary

The existing ownership boundary remains unchanged.

External Server Detected and Adopted External Server are not app-managed processes.

MLX Server Manager does not take ownership of external servers, stop them, restart them, kill them, monitor their memory, or collect their logs.

Verification

• Debug build succeeded
• Release zip created
• Onboarding Guidance panel added
• State-specific guidance added
• Start / Stop / Restart safety unchanged
• Adopt / Forget behavior unchanged
• Advanced Launch Options unchanged
• No inference proxy added
• No chat UI added
• No model downloader added

Asset

MLXServerManager-v2.4.0-unsigned.zip

SHA-256:

ad91d6c9794a496acd44ec4fdf89c89f02e4020169978249b33518c270b75fa5

Size: 361K

v2.3.0 — First-run Onboarding Guidance Design

v2.3.0

This is a docs-only release.

Summary

v2.3.0 adds the onboarding and first-run guidance design for new MLX Server Manager users.

The goal is to make the first setup flow clearer, especially around mlx_lm.server executable paths, model profiles, readiness checks, connection settings, and the difference between managed and external server workflows.

No app code was changed in this release.

Added

• Added docs/onboarding_first_run.md
• Added onboarding / first-run guidance link to README
• Added future onboarding verification notes to docs/testing.md
• Updated task notes for v2.3.0

First-run Guidance Covers

The new design document covers:

• first-run checklist
• required setup
• recommended first launch flow
• managed server flow
• external server flow
• Connection Settings flow
• Hermes Agent setup flow
• troubleshooting prompts
• UI copy suggestions
• safety and privacy notes
• testing plan
• future work

Key First-run Concepts

The guidance explains:

• how to set the mlx_lm.server executable path
• how to confirm the selected model profile
• how host and port are used
• why readiness uses /v1/models
• what to copy for OpenAI-compatible clients
• how Hermes Agent configuration fits into Direct Mode
• how managed server and adopted external server workflows differ
• why the app does not proxy inference requests

Direct Mode Boundary

Direct Mode remains unchanged:

OpenAI-compatible client -> mlx_lm.server or adopted external server -> MLX model

MLX Server Manager does not proxy inference requests, act as a chat UI, or become a multi-backend wrapper.

Ownership Boundary

The guidance keeps the ownership boundary clear:

• Managed Server is started and controlled by MLX Server Manager
• External Server Detected is not app-managed
• Adopted External Server is connection context only
• Stop / Restart applies only to app-managed processes
• External servers are not stopped, restarted, killed, monitored, or log-collected by MLX Server Manager

Not Changed

• No Swift changes
• No Xcode project changes
• No build setting changes
• No new app binary
• No new zip asset
• No inference proxy
• No chat UI
• No multi-backend wrapper
• No external process ownership changes

Current Binary Asset

The current downloadable app binary remains the v1.9.0 unsigned build:

MLXServerManager-v1.9.0-unsigned.zip

This v2.3.0 release does not include a new binary asset.

v2.2.0 — README Screenshot Refresh

v2.2.0

This is a docs/public polish release.

Summary

v2.2.0 refreshes the README screenshots for the current MLX Server Manager UI.

No app code was changed in this release.

Added

• Added updated main dashboard screenshot
• Added Connection Settings / Current Target summary screenshot
• Added Adopted External Server screenshot
• Updated the README Screenshots section
• Updated screenshot refresh notes
• Updated task notes for v2.2.0

Screenshots Added

• screenshots/main-dashboard-v2.2.png
• screenshots/connection-settings-current-target-v2.2.png
• screenshots/adopted-external-server-v2.2.png

Screenshot Coverage

The updated README screenshots now cover:

• Main dashboard
• Model profile list
• Settings and diagnostics
• Server status
• Selected model details
• OpenAI-compatible connection settings
• Current Target summary
• Copy actions for client setup
• Adopted External Server state
• Connection context only behavior
• External server ownership boundary

Direct Mode Boundary

Direct Mode remains unchanged:

OpenAI-compatible client -> mlx_lm.server or adopted external server -> MLX model

MLX Server Manager still does not proxy inference requests, act as a chat UI, or become a multi-backend wrapper.

Ownership Boundary

The Adopted External Server screenshot shows the intended boundary:

• connection context only
• not managed by MLX Server Manager
• Forget External Server action
• Stop / Restart unavailable for external server contexts

MLX Server Manager does not take ownership of external processes.

Privacy

The screenshots were reviewed to avoid exposing:

• real API keys
• tokens
• private paths
• personal home paths
• private repository URLs

Not Changed

• No Swift changes
• No Xcode project changes
• No build setting changes
• No new app binary
• No new zip asset
• No inference proxy
• No chat UI
• No multi-backend wrapper

Current Binary Asset

The current downloadable app binary remains the v1.9.0 unsigned build:

MLXServerManager-v1.9.0-unsigned.zip

This v2.2.0 release does not include a new binary asset.

v2.1.0 — Screenshot Refresh Design

v2.1.0

This is a docs-only release.

Summary

v2.1.0 adds the screenshot refresh design and capture plan for the current MLX Server Manager UI.

No app code was changed in this release.

Added

• Added docs/screenshot_refresh.md
• Added a README link to the screenshot refresh design
• Updated task notes for v2.1.0

Screenshot Plan

The new screenshot refresh design covers the recommended screenshot set for the current UI:

• Main dashboard
• Connection Settings / Current Target summary
• External Server Detected
• Adopted External Server
• Advanced Launch Options
• Logs / diagnostics panel

Capture Scenarios

The design defines capture scenarios for:

• Managed Server Running
• Connection Settings Current Target
• External Server Detected
• Adopted External Server
• Advanced Launch Options
• Logs / Diagnostics

Privacy Checklist

The screenshot plan includes a privacy and redaction checklist to avoid exposing:

• real API keys
• Hugging Face tokens
• GitHub tokens
• private paths
• personal home paths
• private repository URLs
• local shell history

Recommended visible defaults include:

• host: 127.0.0.1
• port: 8080
• API key placeholder: not-required-local

Not Changed

• No Swift changes
• No Xcode project changes
• No build setting changes
• No new app binary
• No new zip asset
• No new screenshot image files
• No README links to missing screenshot files
• No inference proxy
• No chat UI
• No multi-backend wrapper

Current Binary Asset

The current downloadable app binary remains the v1.9.0 unsigned build:

MLXServerManager-v1.9.0-unsigned.zip

This v2.1.0 release does not include a new binary asset.

v2.0.0 — Public README and Docs Polish

v2.0.0

This is a docs-only release.

Summary

v2.0.0 refreshes the public README and related documentation to reflect the current MLX Server Manager feature set through v1.9.0.

No app code was changed in this release.

Updated

• Updated the README introduction
• Added a clearer “What This Is” section
• Refined the “What This Is Not” boundary
• Updated the feature list for the v1.9.0 feature set
• Updated Direct Mode wording
• Updated docs links
• Updated distribution and testing docs to reference the current v1.9.0 binary
• Updated External Server Detection, Adopt External Server, and Connection Settings docs to match the current implementation state
• Updated task notes for v2.0.0

Current Feature Set

The README now reflects the current feature set:

• Start / Stop / Restart managed mlx_lm.server
• Readiness checks via /v1/models
• Managed server logs and memory visibility
• Model profiles
• Advanced Launch Options
• External Server Detection
• Adopt External Server
• Forget External Server
• Connection Settings Current Target summary
• Copy actions for Base URL, Model ID, API key placeholder, JSON config, Hermes Agent config, and /v1/models readiness checks

Direct Mode Boundary

Direct Mode remains unchanged:

OpenAI-compatible client -> mlx_lm.server or adopted external server -> MLX model

MLX Server Manager does not proxy inference requests, act as a chat UI, or become a multi-backend wrapper.

Ownership Boundary

External Server Detection and Adopt External Server documentation now better reflects the current behavior.

Adopted External Server remains connection context only. MLX Server Manager does not take ownership of external processes, stop them, restart them, kill them, monitor their memory, or collect their logs.

Not Changed

• No Swift changes
• No Xcode project changes
• No build setting changes
• No new app binary
• No new zip asset
• No inference proxy
• No chat UI
• No multi-backend wrapper
• No external process ownership changes

Current Binary Asset

The current downloadable app binary remains the v1.9.0 unsigned build:

MLXServerManager-v1.9.0-unsigned.zip

This v2.0.0 release does not include a new binary asset.

v1.9.0 — Connection Settings Polish

v1.9.0

Summary

v1.9.0 improves the Connection Settings UI and copy workflow.

This release adds a Current Target summary so users can clearly understand which server target they are configuring across Managed Server, External Server Detected, Adopted External Server, and Not Running states.

Added

• Added Current Target summary to Connection Settings
• Added Target Type display
• Added ownership note display
• Added readiness/status summary
• Added Copy Base URL
• Added Copy Model ID
• Added Copy API Key Placeholder
• Added Copy All Connection Settings
• Added Copy Hermes Agent Config
• Added Copy curl Readiness Check
• Added ConnectionTargetSummary
• Updated README and docs for the v1.9.0 Connection Settings polish

Target Types

Connection Settings now distinguishes:

• Managed Server
• External Server Detected
• Adopted External Server
• Not Running / Not Connected

For Not Running / Not Connected, the UI shows a preview-style state so users can see the selected configuration without confusing it for an active connection target.

Copy Actions

The Connection Settings UI now provides clearer copy actions for common client setup workflows:

• Base URL
• Model ID
• API key placeholder
• JSON config
• Hermes Agent config
• All connection settings
• curl readiness check
• OpenAI-compatible chat example

These are copy-only helpers. MLX Server Manager does not send inference requests.

Hermes Agent Config

The Hermes Agent config copy text includes:

• Base URL
• Model
• API key placeholder
• Direct Mode note
• local server note
• ownership note
• Qwen thinking note where relevant

Readiness Check

The readiness copy action uses /v1/models.

MLX Server Manager does not use /v1/chat/completions for readiness or detection.

Direct Mode Boundary

Direct Mode remains unchanged:

OpenAI-compatible client -> mlx_lm.server or adopted external server -> MLX model

MLX Server Manager still does not proxy inference requests, act as a chat UI, or become a multi-backend wrapper.

Ownership Boundary

The existing ownership boundary remains unchanged.

External Server Detected and Adopted External Server are not app-managed processes.

MLX Server Manager does not:

• take PID ownership of external servers
• stop external servers
• restart external servers
• kill external servers
• monitor external server memory
• collect external server logs

Managed server lifecycle controls remain limited to processes started by MLX Server Manager.

Verification

• Debug build succeeded
• Release zip created
• Connection Settings target summary added
• Copy actions verified at build level
• Stop / Restart safety unchanged
• /v1/models remains the readiness endpoint
• No inference proxy added
• No chat UI added
• No external process ownership added

Asset

MLXServerManager-v1.9.0-unsigned.zip

SHA-256:

8bf679d1252d444def26a95a1c1bcb31c42cc9c4fe2401afe79ec3e68d6b5696

Size: 351K