Laboratory

Archived MVP · Retired August 2026

Inspect the whole experiment.

LM Laboratory was a SwiftUI workbench for connecting local model providers, running inspectable chats, assembling personas and teams, generating images, and seeing the prompts, payloads, settings, and failures that other interfaces hide. Product work has ended; the repository remains as a technical reference.

Review the archive notes ↓Private source ↗
GitHub access is required to view this private repository.
WORKBENCH / LIVE ROUTE
PROVIDERLM Studiodiscover · load · stream
OUTPUTInspectableprovenance · timing · state
LOCAL-FIRSTSWIFTDATAKEYCHAIN
PlatformSwiftUI + SwiftData
Last substantive workMay 2026
Project stateArchived / retired
SourcePrivate repository

Implemented bench

A real core, beyond the initial shell.

The source and focused tests show a working local-model laboratory. Provider breadth and Apple system integrations remain narrower than the PRD's long-range vision.

01

Connections & models

Persist LM Studio and image-server connections, discover models, run health checks, and use the native load/unload protocol.

DISCOVERY · LIFECYCLE · SECURITY POLICY
02

Inspectable chat

Stream LM Studio responses through native or OpenAI-compatible routes, preserve chat history, and expose run settings and status blocks.

STREAMING · HISTORY · PROVENANCE
03

Personas & teams

Persist structured personas, model assignment, prompts, member order, and manual or round-robin team turns.

PROMPTS · ROUTING · TEAM STATE
04

Forge image work

Discover A1111/Forge capabilities, construct advanced requests, generate outputs, and keep image bytes on disk with metadata in SwiftData.

CAPABILITIES · GENERATION · FILE STORAGE
05

Privacy & portability

Store credentials through Keychain, reject unsafe public HTTP, redact diagnostics, and support encrypted export without secrets.

KEYCHAIN · REDACTION · EXPORT
06

Honest boundary

Cloud image APIs, on-device image generation, App Intents, Live Activities, benchmarking, and sync are direction—not completed capabilities.

PLANNED · NOT SHIPPED

Lab notebook / archive

Retired as a product. Retained as a reference.

  1. BuiltFoundation

    Functional native MVP

    Persistent connections, models, chat history, Personas, Teams, settings, LM Studio operations, Forge generation, diagnostics, encrypted data transfer, and focused unit/UI coverage.

  2. RetiredAugust 2026

    Product work has ended

    The App Store record was retired and there is no active release roadmap. This repository remains intact as evidence of the work and as a source of reviewed patterns for active products.

  3. Retained valueReference only

    Carry forward the strongest patterns

    Inspectable prompts and payloads, Keychain-backed credentials, redacted diagnostics, encrypted transfer, and compact provider workflows can inform LM Manager Pro, Mira, or ComfyChat without reviving a duplicate product.

  4. Restart ruleNo active plan

    Require a distinct product reason

    Any future return would begin with a new product decision, a clearly differentiated user problem, and fresh build and release validation—not with the old roadmap.

Project lineage

A sibling, not a fork.

Its git history is independent, but its product territory overlaps three native AI studio experiments. That makes LM Laboratory useful as a comparison point: it emphasizes inspectability and a compact, conventional MVVM workbench.

LM Manager ProBroader and much more mature creation platform.Lucidity / MiraCleaner chat-first shell with a separate self-hosted Fleet.ComfyChatFocused local chat + durable ComfyUI workflow product.