🧠 LCM Explorer

A web UI for browsing and searching OpenClaw LCM (Lossless Context Management) databases.

Browse conversations, drill into messages and their parts, explore the summary DAG, search across everything with full-text search, and inspect the context window that your AI actually sees.

Quick Start

npx lcm-explorer

Opens on http://localhost:3078. Auto-discovers your LCM database from your OpenClaw config, or falls back to ~/.openclaw/lcm.db.

If the database isn't found, you'll get specific guidance on what to do next (not just an error).

What is LCM?

lossless-claw is a context engine plugin for OpenClaw that replaces the default sliding window with lossless context management. It stores every message verbatim in a SQLite database and builds a DAG of summaries so your AI never truly forgets a conversation.

LCM Explorer gives you a window into that database.

Prerequisites

• Node.js 18+ (run node --version to check)
• An LCM database from a running OpenClaw instance with the lossless-claw plugin

Don't have LCM set up yet? See Troubleshooting below.

Tip: Ask your OpenClaw bot to install and set up LCM Explorer for you. It already knows your config, whether LCM is enabled, and where the database lives. Just tell it: "Install LCM Explorer and set it up so I can browse my conversation history." It'll handle the rest, including enabling the LCM plugin if needed.

Install

npx (no install needed)

npx lcm-explorer

npm global

npm install -g lcm-explorer
lcm-explorer

Docker

docker run -v ~/.openclaw:/data -p 3078:3078 ghcr.io/sheevn/lcm-explorer

From source

git clone https://github.com/Sheevn/LCM-Explorer.git
cd LCM-Explorer
npm install
npm start

Configuration

CLI flags

lcm-explorer [options]

  --port, -p <port>    Server port (default: 3078)
  --db <path>          Path to lcm.db (overrides auto-discovery)
  --help, -h           Show help
  --version, -v        Show version

Environment variables

| Variable | Default | Description | |----------|---------|-------------| | PORT | 3078 | Server port | | LCM_DB_PATH | (auto-discover) | Explicit path to LCM SQLite database | | OPENCLAW_HOME | ~/.openclaw | OpenClaw home directory (for config discovery) | | AUTH_TOKEN | (none) | Optional access token for authentication |

Database discovery

LCM Explorer finds your database automatically:

1. --db flag or LCM_DB_PATH env var: explicit path, used as-is
2. OpenClaw config: reads openclaw.json to find the LCM plugin's configured DB path
3. Default: ~/.openclaw/lcm.db

For non-standard OpenClaw installs, set OPENCLAW_HOME:

OPENCLAW_HOME=/opt/openclaw npx lcm-explorer

Custom database path

# Via flag
npx lcm-explorer --db /path/to/lcm.db

# Via env var
LCM_DB_PATH=/path/to/your/lcm.db npx lcm-explorer

Custom port

npx lcm-explorer --port 9090
# or
PORT=9090 npx lcm-explorer

Authentication

By default there's no auth, which is fine for local use.

If you're exposing it on a network, set AUTH_TOKEN:

AUTH_TOKEN=my-secret-token npx lcm-explorer

You'll get a login page. Enter the token once and you're in for 7 days (cookie-based session).

API access works with Bearer tokens:

curl -H "Authorization: Bearer my-secret-token" http://localhost:3078/api/stats

Features

• Conversations: Browse all conversations with message counts, token usage, and timestamps
• Messages: Paginated message list with role filtering, click through to full content and parts
• Summary Tree: Visualize the summary DAG with parents, children, and source messages
• Full-Text Search: Search across messages and summaries using SQLite FTS5
• Context Window: See exactly what items are in the current context window
• Statistics: Database overview with role distribution, compression ratios, part type breakdown
• Dark/Light Theme: Toggle between themes, preference saved locally

Screenshots

Troubleshooting

"LCM database not found"

LCM Explorer gives you specific guidance based on your setup:

• No OpenClaw installed: Install OpenClaw first, then the LCM plugin
• OpenClaw installed, no LCM plugin: Run openclaw plugins install @martian-engineering/lossless-claw
• LCM plugin installed, not configured: Add it as a contextEngine slot in openclaw.json
• LCM configured, no data yet: Run some conversations with OpenClaw to generate data

You can always point directly to a database:

npx lcm-explorer --db /path/to/lcm.db

"Port 3078 is already in use"

Another process is using that port. Try a different one:

npx lcm-explorer --port 3079

Database on a remote server

Copy the database locally (it's a single SQLite file):

scp server:~/.openclaw/lcm.db ./lcm.db
npx lcm-explorer --db ./lcm.db

The database is opened read-only, so it's safe to copy while OpenClaw is running.

LCM Compatibility

| LCM Version | Explorer Support | Notes | |-------------|-----------------|-------| | 0.3.x | Full | Core tables only | | 0.4.x | Full | Plugin config support, session scoping | | 0.5.0+ | Full | Large file externalization, bootstrap checkpoints, model on summaries |

LCM Explorer auto-detects which tables exist and degrades gracefully. Tables added in newer versions (like large_files and conversation_bootstrap_state) are queried only if present. Running against an older LCM database works fine; you just won't see the newer stats.

API Routes

All routes return JSON. Authenticated if AUTH_TOKEN is set.

| Route | Description | |-------|-------------| | GET /health | Always public. Returns status, uptime, DB connection state | | GET /api/stats | Aggregate stats: messages, summaries, conversations, role distribution, large files, bootstrap checkpoints, DB size | | GET /api/conversations | List all conversations with message/summary counts and token totals | | GET /api/conversations/:id | Single conversation detail with role breakdown | | GET /api/conversations/:id/messages | Paginated messages. Params: limit, offset, role | | GET /api/conversations/:id/messages/:msgId | Full message with all parts | | GET /api/conversations/:id/summaries | Summaries with parent IDs and source message IDs | | GET /api/summaries/:id/tree | Summary DAG: parents, children, source messages | | GET /api/context/:convId | Current context window items with resolved content | | GET /api/search | Full-text search. Params: q, scope, conversation_id, limit | | GET /api/large-files | List externalized large files (v0.5.0+). Params: conversation_id, limit, offset | | GET /api/large-files/:id | Single file detail (v0.5.0+) |

LCM Database Schema Reference

Tables created and managed by lossless-claw. LCM Explorer reads these in read-only mode.

Core Tables (all versions)

| Table | Purpose | Key Columns | |-------|---------|-------------| | conversations | Session groupings | conversation_id, session_id, session_key, title, bootstrapped_at, created_at, updated_at | | messages | Raw message content | message_id, conversation_id, seq, role, content, token_count, created_at | | message_parts | Structured message components | part_id, message_id, session_id, part_type, ordinal, text_content, tool_name, tool_input, tool_output, tool_error, tool_call_id, tool_status, tool_title, patch_hash, patch_files, file_mime, file_name, file_url, subtask_prompt, subtask_desc, subtask_agent, step_reason, step_cost, step_tokens_in, step_tokens_out, snapshot_hash, compaction_auto, metadata, is_ignored, is_synthetic | | summaries | Compressed context (DAG nodes) | summary_id, conversation_id, kind (leaf/condensed), depth, content, token_count, earliest_at, latest_at, descendant_count, descendant_token_count, source_message_token_count, file_ids, model, created_at | | summary_messages | Summary-to-message links | summary_id, message_id, ordinal | | summary_parents | DAG edges | summary_id, parent_summary_id, ordinal | | context_items | Current context window | conversation_id, ordinal, item_type (message/summary), message_id, summary_id, created_at |

v0.5.0+ Tables

| Table | Purpose | Key Columns | |-------|---------|-------------| | large_files | Externalized tool outputs | file_id, conversation_id, file_name, mime_type, byte_size, storage_uri, exploration_summary, created_at | | conversation_bootstrap_state | Incremental bootstrap checkpoints | conversation_id, session_file_path, last_seen_size, last_seen_mtime_ms, last_processed_offset, last_processed_entry_hash, updated_at |

FTS Tables

| Table | Tokenizer | Indexed Columns | |-------|-----------|-----------------| | messages_fts | porter unicode61 | content | | summaries_fts | porter unicode61 | summary_id (UNINDEXED), content |

Message Part Types

text, reasoning, tool, patch, file, subtask, compaction, step_start, step_finish, snapshot, agent, retry

Tech

• Express + better-sqlite3 (read-only)
• Vanilla JS SPA (no build step, no framework)
• CSS custom properties for theming
• Zero external API calls

License

MIT