feat: first-class OpenClaw plugin for Traul as agent external memory

[dandaka]

Vision

Make Traul a first-class plugin in the OpenClaw ecosystem so any OpenClaw user can install Traul and use it as external memory for their AI agent. The setup should be straightforward — ideally a one-liner install from ClawHub.

Why This Matters

OpenClaw has 135K+ GitHub stars and 700+ skills in its marketplace. It already connects to the same messaging platforms Traul syncs from (WhatsApp, Telegram, Discord, Slack, etc.). By positioning Traul as OpenClaw's memory layer, we tap into a massive user base that already has the messaging infrastructure set up — they just need the search/intelligence layer on top.

What Traul Brings to OpenClaw Users

• Cross-platform message search — hybrid FTS5 + vector search across all connected platforms
• Signal detection — SQL-based pattern matching (stale threads, etc.)
• Briefings — structured daily overviews with volume charts
• Local-first — no data leaves the user's machine
• Agent-native — CLI with JSON output, designed to be called by AI agents

Implementation Plan

Phase 1: MCP Server Mode

• Add an MCP server transport to Traul (stdio or HTTP)
• Expose core tools via MCP: traul_search, traul_messages, traul_channels, traul_signals, traul_briefing, traul_sql
• This alone makes Traul usable by any MCP-compatible agent (Claude, OpenClaw, etc.)

Phase 2: OpenClaw Native Plugin

• Create openclaw.plugin.json manifest
• Package as npm module (@traul/openclaw-plugin or openclaw-plugin-traul)
• Register tools in OpenClaw's capability registry
• Add background service for periodic sync (leverage OpenClaw's Gateway lifecycle)
• Handle config bridging — reuse OpenClaw's existing platform credentials where possible (avoid double-auth)

Phase 3: ClawHub & Distribution

• Publish to ClawHub marketplace for one-click install
• Write setup guide: openclaw plugins install traul
• Add OpenClaw-specific README section with screenshots/examples
• Create demo skill that shows an agent using Traul to answer "what did X say about Y last week?"

Phase 4: Deep Integration

• Auto-discover OpenClaw's connected platforms and offer to sync them
• Share contact/identity data with OpenClaw's contact system
• Expose signals as OpenClaw notifications
• Support OpenClaw's RPC interface for real-time queries from agents

Technical Notes

• OpenClaw plugins are TypeScript modules loaded via jiti at runtime
• Plugins can register: tools, CLI commands, RPC methods, HTTP routes, background services
• MCP servers from plugin bundles run as subprocesses — Traul's MCP mode could work this way
• OpenClaw already has memory plugins (Core/LanceDB) — Traul differentiates by being communication-focused with multi-source sync, not general-purpose memory
• Plugin security: installed with npm install --ignore-scripts, path security checks apply

Open Questions

• Should Traul's sync run inside the OpenClaw Gateway process or as a separate daemon?
• How to handle credential sharing — can Traul reuse OpenClaw's Telegram/Discord/Slack sessions?
• Should we support OpenClaw's Claude-compatible plugin format (.claude-plugin/plugin.json) as well?
• Naming: traul, openclaw-traul, @traul/openclaw?

References

• OpenClaw Plugin Docs
• ClawHub Marketplace
• OpenClaw GitHub
• Current Traul skill.md already defines the tool interface — this is the starting point for MCP tools