OpenClaw Gateway & Agent Stack
80+ containers — one per user
Telegram → OpenClaw → skills
Each container runs independently
Dynamic skill loading (Python/JS)
80+ containers → high ops cost
Individual Agents
Each user has isolated agent instance
No cross-agent memory or coordination
Hard to propagate updates
Isolated per user
Hybrid gBrain Knowledge Graph
Vector + BM25 + Graph search
Entity recognition & linking
Stored per-agent in SQLite
No centralized learning
Per-agent only
6 Workflow Archetypes
Action / Coach / Expert / Planner / Monitor / Profile
Runs inside each agent
Hard to version control
⚡ Action20 skills
🫂 Coach30 skills
🎓 Expert7 skills
🗺 Planner10 skills
👁 Monitor8 skills
👤 Profile3 skills
Same across all
Skill Engine API
Central logic: routing, deep profile, journeys
Called by iOS app and OpenClaw agents
Python backend (sqlite, OpenRouter)
Used by both
🚀
HERMES + OiMy Integration
HERMES Agent Gateway
Single cluster for all users
92K★ TypeScript engine, compiled
Telegram routing → user profile lookup
Replaces all OpenClaw containers
1 cluster → low ops cost
23 Compiled Workflows
TypeScript + compiled binary
Pre-verified, type-safe, secure
Faster than dynamic skill loading
Replaces per-agent workflow layer
Replace OpenClaw scripts
Centralized Hybrid gBrain
Global knowledge graph across all users
Pattern detection → cross-family insights
Entity disambiguation at scale
Preserves & enhances per-agent gBrain
Cross-user learning
Archetype Workflow Layer
Wraps 6 archetypes as workflows
Action → workflow → zero token
Coach → workflow → template or tall model
Preserves full OiMy archetype system
Action
Coach
Expert
Planner
Monitor
Profile
Structured execution
Persistent Browser State
Auto-login, stateful sessions
Perfect for web automation (banking, email, social)
No repeated login screens
Fully new capability
New superpower
Skill Engine API (shared)
Central logic: routing, deep profile, journeys
Called by iOS app, HERMES, web app
Identical backend to current system
Used by both