LM Manager Pro

LM Manager Pro is a native iOS and iPadOS app for developers and AI enthusiasts who run their own language model infrastructure. Connect to LM Studio inference servers on your local network or Tailscale VPN, integrate leading cloud providers, and manage everything from one place.

**INFERENCE NODE MANAGEMENT**
Add Mac Minis, workstations, or Tailscale peers as inference nodes. See which models are loaded across all machines at a glance. Load and unload GGUF and MLX models with a single tap, and watch VRAM usage in real time. Background polling keeps status current while the app is open, and the status bar shows network health without interrupting your workflow.

**AI PERSONAS — YOUR ENGINEERING TEAM**
Give each model a human-style identity with a name, role, personality overlay, and custom system prompt. Rank personas from Untested to Expert as they accumulate runs. Assign GPT-4o to your Architect, Claude Opus to your Code Reviewer, Gemini Flash to your Sprint Planner — then put them to work together.

**TEAM CHAT & GROUP RUNS**
Assemble personas into Teams and start a group chat with auto round-robin turn management or manual persona selection. Every team member responds in sequence, giving you parallel perspectives on a single prompt without switching windows. Enable Cross-Talk so team members can reference each other's responses, or disable it for independent answers.

**STRUCTURED EPISODES**
Write reusable, multi-step benchmark scripts with prompt sections, pause points, and notes. Run an episode against a team and watch each section execute in sequence. Every completed run is archived with full response history so you can compare results across models and sessions.

**CROSS-PROVIDER STREAMING**
Stream responses from LM Studio, OpenAI, Anthropic, Google, and Qwen from one interface. Reasoning models can show their thinking content alongside the final response, and vision-capable models accept images from your photo library.

**COMPARE MODE**
Run the same prompt across two to four models in parallel and watch them stream side by side. Live metrics — tokens per second, time-to-first-token, total completion time — update as each model responds. Every comparison is saved automatically so you can revisit past results anytime.

**TIMELINE**
Browse a chronological feed of every AI response across all sessions. Search by content or sender, sort by newest, oldest, fastest TPS, or most tokens, and filter by session type or model. Cloud and local responses are color-coded so you can scan at a glance.

**AI-SCORED LEADERBOARD**
Automatically score each persona across five dimensions — Accuracy, Helpfulness, Instruction Following, Clarity, and Depth — using an AI advisor model of your choice. Track performance over time and compare ranked standings on the Leaderboard.

**VOICE INPUT & OUTPUT**
Dictate messages with on-device speech recognition and hear responses read back with configurable voices.

**MEMORY & CONTEXT**
Enable per-persona memory, override memory per session, and use DuckDuckGo-powered web search when a persona needs current context.

**YOUR DATA, YOUR DEVICE**
All data is stored locally in SwiftData — no accounts, no telemetry, no cloud sync required. Export your full dataset as JSON for backup or device migration. Import it back with one tap.

**BUILT FOR POWER USERS**
- Side-by-side model evaluation with saved history
- Token usage and latency metrics for every response
- Timeline search, sort, and filters
- Markdown and code block rendering
- Reusable prompt templates
- Full chat export as plain text or Markdown
- Projects for organizing sessions, teams, and personas