OpenClaw as a Multi-User Telegram Bot Engine

1. OpenClaw Architecture for Multi-User

How It Works

OpenClaw is an AI Gateway that receives messages from channels (Telegram, WhatsApp, Discord, etc.) and routes them to agents. Key abstractions:

Session Model

OpenClaw supports several session.dmScope:

Multi-Agent Routing

You can run multiple agents in a single Gateway:

{
  agents: {
    list: [
      { id: "protech", workspace: "~/protech-workspace", default: true },
      { id: "admin", workspace: "~/.openclaw/workspace" },
    ],
  },
  bindings: [
    // Fedor → admin agent (full access)
    { agentId: "admin", match: { channel: "telegram",
      peer: { kind: "direct", id: "224153441" } } },
    // All others → protech agent (restricted)
    { agentId: "protech", match: { channel: "telegram" } },
  ],
}

This means: Fedor talks to the full-featured Jarvis, while all other users interact with a restricted ProTechBot agent.

Lane-Aware Queue

Messages are processed through a FIFO queue with a concurrency cap:

• • Main lane: default concurrency = 4
• • Subagent lane: default concurrency = 8
• • Configurable: agents.defaults.maxConcurrent

2. Hardware Requirements

Mac Mini M4 Pro (24GB RAM, 512GB SSD)

OpenClaw Gateway is a Node.js process. Main resource consumption:

Scenario Calculations

Important: OpenClaw does NOT run LLMs locally — it sends requests to API providers. The main load is therefore network + session state.

Вердикт: Mac Mini M4 Pro 24GB потянет 500 users по ресурсам, но bottleneck будет в concurrency обработки и скорости LLM API.

3. Session Management

Context Storage

Sessions are stored on disk:

~/.openclaw/agents/<agentId>/sessions/sessions.json  — индекс
~/.openclaw/agents/<agentId>/sessions/*.jsonl         — полная история

Format is JSONL (JSON Lines): each turn (user/assistant/tool) is a separate line. Files are never modified, only appended to (append-only).

Memory per User

OpenClaw has two memory levels:

Solutions for Personal Memory

~/protech-workspace/
  users/
    tg_123456789/
      profile.md
      history.md
    tg_987654321/
      profile.md
      history.md
  shared/
    knowledge-base/

4. Shared Knowledge Base + Personal Memory

RAG Architecture on OpenClaw

OpenClaw doesn't have built-in RAG. But there are several approaches:

5. Security and Isolation

Prompt Injection Protection

OpenClaw has multi-layered protection:

User-to-User Isolation

Risks

6. Admin Access

How the Admin Manages the Bot

Monitoring

# Логи сессий
~/.openclaw/agents/<agentId>/sessions/*.jsonl

# Gateway логи
openclaw logs --tail

# Security audit
openclaw security audit --deep

# Health check
openclaw health

7. Scaling

Current Limitations

Scaling Paths

8. Cost Model

LLM API Costs per User

Average ProTechBot user: ~5-20 messages/day, ~500-2000 tokens per turn.

Additional Costs

9. Comparison with Alternatives

Detailed Breakdown

10. Real-World Community Experience

GitHub Issues (OpenClaw)

Key Community Takeaways

11. Verdict and Recommendations

Can OpenClaw Be Used for ProTechBot?

Short answer: YES, but with caveats.

Pros of OpenClaw for ProTechBot

Cons of OpenClaw for ProTechBot

Recommended Architecture: Hybrid Approach

Конфигурация для чистого OpenClaw (MVP до 50 users)

{
  session: {
    dmScope: "per-channel-peer",
    reset: { mode: "idle", idleMinutes: 120 },
  },
  agents: {
    defaults: {
      model: { primary: "deepseek/deepseek-chat" },
      maxConcurrent: 10,
      contextTokens: 100000,
    },
    list: [
      {
        id: "protech",
        default: true,
        workspace: "~/protech-bot/workspace",
        sandbox: { mode: "all", scope: "session", workspaceAccess: "ro" },
        tools: {
          allow: ["read", "web_search", "session_status"],
          deny: ["write", "edit", "exec", "process", "browser",
                 "nodes", "canvas"],
        },
      },
      {
        id: "admin",
        workspace: "~/.openclaw/workspace",
      },
    ],
  },
  bindings: [
    { agentId: "admin", match: { channel: "telegram",
      peer: { kind: "direct", id: "224153441" } } },
  ],
  channels: {
    telegram: {
      dmPolicy: "open",
    },
  },
}

Roadmap

Appendices

A. Useful Configuration Commands

# Настроить dmScope
openclaw config set session.dmScope "per-channel-peer"

# Открыть DM для всех
openclaw config set channels.telegram.dmPolicy "open"

# Увеличить concurrency
openclaw config set agents.defaults.maxConcurrent 10

# Security audit
openclaw security audit --deep

# Посмотреть активные сессии
openclaw sessions --active 60

# Логи
openclaw logs --tail --level info

B. References