lm-assist

Claude Code Mission Control -- Monitor sessions, track costs, manage knowledge, from anywhere.

lm-assist is a knowledge management and session inspection tool for Claude Code. It captures what you learn from every coding session and automatically injects that knowledge into future prompts -- so Claude never re-explores the same codebase twice.

Key Capabilities

How It Works

Claude Code Session
       |
       v
  lm-assist captures session data
       |
       +----> Knowledge Base (auto-extracted facts)
       |           |
       |           v
       |      MCP Tools (search, detail, feedback)
       |           |
       |           v
       |      Context Injection Hook
       |           |
       |           v
       +----> Next Claude Code Session (knowledge pre-loaded)

Without lm-assist: Each new session starts from scratch, re-exploring the same codebase (~85k tokens, ~3 minutes of overhead).

With lm-assist: Knowledge from previous sessions is automatically injected (~3k tokens, < 1 second). 95%+ reduction in exploration overhead.

Quick Start

In Claude Code:

# Add the LangMart marketplace
/plugin add-marketplace langmartai/claude-plugins-official

# Install lm-assist
/plugin install lm-assist

# Open dashboard
/assist

See Installation for alternative install methods (npm, from source).

Screenshots

Session Browser

Browse all Claude Code sessions with token counts, costs, and model information.

Session Detail (13 Tabs)

Full conversation timeline with syntax-highlighted code blocks.

Agent Tree

Visualize subagent hierarchy -- Explore, Plan, Bash, and custom agents.

Knowledge Base

Browse, search, and manage auto-extracted knowledge entries.

Task Kanban

Aggregated task board from all active sessions.

Web Terminal

Live terminal access to running Claude Code sessions from any browser.

Architecture -- Local / LAN / Cloud Hybrid

lm-assist operates in three access modes that work simultaneously. Your data always stays on your machine -- cloud mode only relays API requests, never stores your session data.

Local Components

Three Access Modes

                     +-----------------------+
                     |   Your Dev Machine    |
                     |                       |
                     |  Claude Code          |
                     |    |                  |
                     |    v                  |
                     |  lm-assist            |
                     |   Core API (:3100)    |
                     |   Web UI   (:3848)    |
                     |   Terminals (:5900+)  |
                     +-----------+-----------+
                                 |
              +------------------+------------------+
              |                  |                  |
              v                  v                  v
        LOCAL MODE          LAN MODE          CLOUD MODE
     localhost:3848      192.168.x.x:3848    langmart.ai
     No auth needed      Optional LAN auth   API key auth
     Same machine        Same network         Any device,
                         Phone, tablet        anywhere

Local Mode -- Access the dashboard at http://localhost:3848. No authentication required. This is the default.

LAN Mode -- Access from other devices on your local network (e.g., http://192.168.1.100:3848 from your phone or tablet). Optional LAN authentication can be enabled to require a token for network access.

Cloud Mode -- Connect to the LangMart Hub to access your dashboard from anywhere via https://langmart.ai/assist-dashboard. All requests are relayed through the Hub WebSocket connection -- your data never leaves your machine.

Cloud Mode: Remote Terminal Access

The most powerful feature of cloud mode is remote terminal access -- control your Claude Code sessions from any browser, anywhere in the world.

  Browser (anywhere)
       |
       v
  langmart.ai (Hub)
       |
       | WebSocket relay
       | (binary frames)
       |
       v
  lm-assist (your machine)
       |
       v
  ttyd (local terminal)
       |
       v
  Claude Code session

How it works:

1. lm-assist maintains a persistent WebSocket connection to the LangMart Hub
2. When you open a terminal in the cloud dashboard, the Hub sends a console_start_relay request
3. lm-assist starts a local ttyd process for the Claude Code session
4. Binary terminal frames are relayed between your browser and the local ttyd process through the Hub WebSocket
5. The relay uses efficient binary framing with SHA256 hash-based session multiplexing for concurrent sessions

This means you can start a Claude Code session on your desktop, then monitor and interact with it from your phone during your commute -- no VPN or port forwarding required.

Cloud Mode: Multi-Machine Knowledge

When multiple machines connect to the Hub, knowledge is shared across all of them:

  Machine A (Linux server)          Machine B (MacBook)
       |                                 |
       |  lm-assist                      |  lm-assist
       |  Knowledge Base A               |  Knowledge Base B
       |                                 |
       +--------- LangMart Hub ---------+
                       |
                       v
              Unified Knowledge
              (search across all machines)

• Session browsing shows sessions from all connected machines
• Context injection pulls knowledge from your entire fleet
• The Machines page shows online/offline status of all connected instances

Setting Up Cloud Mode

lm-assist setup --key YOUR_LANGMART_API_KEY

Or configure manually:

TIER_AGENT_HUB_URL=wss://api.langmart.ai
TIER_AGENT_API_KEY=sk-your-key-here

Get an API key at langmart.ai/api-keys or use the guided setup at langmart.ai/assist.

Verify with /assist-status in Claude Code -- you should see "Hub: connected".

Access Mode Comparison

Next Steps

• Installation -- Full installation guide with all methods
• Features -- Dashboard walkthrough and feature guide
• Knowledge & MCP Tools -- How the knowledge system works
• GitHub Repository -- Source code and issues

Related Documentation

• API Keys -- Create API keys for lm-assist cloud connection
• Integrations -- Other tool integrations