Memoh-v2

Real Isolation · Real Memory · Real Evolution — The Uncompromising AI Agent Platform

Others share a runtime — we give each bot its own containerd container.

Others use SQLite vector search — we use Qdrant + BM25 + LLM triple extraction.

Others edit memory files by hand — our bots reflect, experiment, and review on their own.

中文 · Feature Guide · Installation · Tutorials · Screenshots

Quick Start

Requires Docker:

curl -fsSL https://raw.githubusercontent.com/Kxiandaoyan/Memoh-v2/main/scripts/install.sh | sh

Silent install (skip prompts): curl -fsSL ... | sh -s -- -y

Or manually:

git clone --depth 1 https://github.com/Kxiandaoyan/Memoh-v2.git
cd Memoh-v2
docker compose up -d

Visit http://localhost:8082. Default login: admin / admin123

After installation, configure in this order:

1. Settings -> Provider    Add API provider, enter API Key and Base URL
       |
2. Provider -> Models      Add models (chat or embedding type)
       |
3. New Bot                 Select a template or start blank, set name and type
       |
4. Bot -> Settings         Choose Chat model, Embedding model, language, etc.
       |
5. Bot -> Channels         Connect Telegram / Lark messaging platforms (optional)

For detailed installation, upgrade, uninstall, and data migration guides, see Installation & Upgrade.

Architecture Overview

                    ┌──────────────┐
                    │   Web UI     │ :8082
                    │  Vue 3       │
                    └──────┬───────┘
                           │
              ┌────────────┼────────────┐
              │            │            │
      ┌───────▼──────┐ ┌──▼──────────┐ │
      │   Server     │ │ Agent       │ │
      │   Go + Echo  │ │ Gateway     │ │
      │   :8080      │ │ Bun + Elysia│ │
      └──┬────┬──────┘ │ :8081       │ │
         │    │         └──┬──────────┘ │
         │    │            │            │
    ┌────▼┐ ┌─▼─────┐  ┌──▼──────────┐ │
    │ PG  │ │Qdrant │  │ Containerd  │◄┘
    │     │ │       │  │ (per-bot    │
    │     │ │       │  │  containers)│
    └─────┘ └───────┘  └─────────────┘

Screenshot Preview

👉 Click for more screenshots

Core Features

Full details for each feature in the Feature Guide.

• Bot Management & Templates — 13 mental-model templates (10 real thought-leaders), two-step professional bot creation
• Conversation & Streaming — SSE real-time streaming + synchronous modes, automatic context management and memory recall
• Three-Layer Memory — Vector semantic search + BM25 keywords + LLM smart extraction, auto-indexed after each turn
• Isolated Container Sandbox — Each bot gets a containerd container with file I/O, shell, browser, and snapshot rollback
• Multi-Platform Channels — Telegram / Lark / Web chat / CLI with cross-platform identity unification
• MCP Tool System — 15 built-in tools + any external MCP server, Stdio and Remote transport
• Heartbeat & Scheduling — Periodic + event-driven dual-mode triggers; bots act proactively
• Self-Evolution — Three-phase organic cycle (Reflect / Experiment / Review) with full evolution log tracking
• Subagents & Skills — Auto-dispatched subagents, one-click skill install from ClawHub marketplace
• OpenViking Tiered Context — L0/L1/L2 structured memory layers, dramatically reducing token consumption
• Token Usage & Diagnostics — Dashboard charts + multi-bot comparison, one-click service health check
• Cross-Bot Collaboration — /shared workspace with file-based coordination

Documentation

Tech Stack

Acknowledgments

This project is a secondary development based on Memoh. Thanks to the original authors for their excellent work.