OpenClaw Small Office Agent Fleet — 5 Agents, 5 Telegram Bots, One Gateway

guide-small-office-agents.md

Small Office Agent Fleet — Example Setup

Five agents, one Gateway, one machine. Each agent has its own Telegram bot so your team can message the right "department" directly.

---

The Agents

📧 Inbox — Email Manager

Monitors email, drafts replies, flags urgent messages, summarizes your inbox every morning.

📱 Social — Social Media Manager

Drafts posts, monitors mentions, suggests engagement replies, tracks what's performing.

📋 Ops — Office Operations

Manages calendars, schedules meetings, tracks tasks, handles internal reminders and SOPs.

🔍 Research — Research Assistant

Digs into topics on demand, summarizes articles, compares vendors, preps briefing docs.

🧾 Books — Bookkeeper

Categorizes expenses, flags anomalies, preps weekly summaries, answers "how much did we spend on X?"

---

The Config

{
  "agents": {
    "defaults": {
      "model": {
        // Use a fast, cheap model as the default
        "primary": "anthropic/claude-sonnet-4-5"
      },
      "heartbeat": { "every": "1h" }
    },
    "list": [
      {
        "id": "inbox",
        "name": "Inbox",
        "workspace": "~/.openclaw/workspace-inbox"
      },
      {
        "id": "social",
        "name": "Social",
        "workspace": "~/.openclaw/workspace-social"
      },
      {
        "id": "ops",
        "default": true,
        "name": "Ops",
        "workspace": "~/.openclaw/workspace-ops"
      },
      {
        "id": "research",
        "name": "Research",
        "workspace": "~/.openclaw/workspace-research",
        // Research gets the big brain
        "model": "anthropic/claude-opus-4-5"
      },
      {
        "id": "books",
        "name": "Books",
        "workspace": "~/.openclaw/workspace-books"
      }
    ]
  },

  "bindings": [
    { "agentId": "inbox",    "match": { "channel": "telegram", "accountId": "inbox" } },
    { "agentId": "social",   "match": { "channel": "telegram", "accountId": "social" } },
    { "agentId": "ops",      "match": { "channel": "telegram", "accountId": "ops" } },
    { "agentId": "research", "match": { "channel": "telegram", "accountId": "research" } },
    { "agentId": "books",    "match": { "channel": "telegram", "accountId": "books" } }
  ],

  "channels": {
    "telegram": {
      "dmPolicy": "pairing",
      "groupPolicy": "allowlist",
      "streamMode": "partial",
      "accounts": {
        "inbox":    { "botToken": "BOT_TOKEN_1" },
        "social":   { "botToken": "BOT_TOKEN_2" },
        "ops":      { "botToken": "BOT_TOKEN_3" },
        "research": { "botToken": "BOT_TOKEN_4" },
        "books":    { "botToken": "BOT_TOKEN_5" }
      }
    }
  }
}

---

Setup Steps

1. Create 5 bots via @BotFather in Telegram (one per agent)
2. Tell your current agent to set up the fleet — give it the names, roles, and tokens
3. Pair each bot — DM each one with /start and approve
4. Customize each workspace — drop a SOUL.md into each with role-specific instructions

---

Example SOUL.md Files

Inbox Agent

# SOUL.md
You are the office email manager. You monitor the inbox, flag urgent
messages, draft replies, and summarize unread mail each morning.

Rules:
- Never send an email without explicit approval
- Flag anything from clients or with "urgent" in the subject
- Morning summary by 8:30 AM

Social Agent

# SOUL.md
You manage the company's social media presence. You draft posts,
track engagement, and suggest replies to comments and mentions.

Rules:
- Never post without approval
- Match the brand voice: professional but approachable
- Flag negative mentions immediately

Books Agent

# SOUL.md
You are the office bookkeeper assistant. You categorize expenses,
track spending trends, and prepare weekly financial summaries.

Rules:
- Never make payments or transfers
- Flag any expense over $500
- Weekly summary every Friday at 4 PM

---

Why This Works

• Focused context. Each agent only sees conversations relevant to its role. The email agent isn't cluttered with social media chatter.
• Right model for the job. Research gets Opus for deep thinking. Everything else runs on Sonnet to keep costs down.
• Team access. Multiple people can pair with each bot. The office manager talks to Ops, the marketing person talks to Social, the accountant talks to Books.
• Isolated memory. Each agent builds its own knowledge over time — the bookkeeper learns your expense categories, the social agent learns your brand voice.
• One machine. All five agents run on a single Gateway process. A Mac Mini or a $20/month VPS handles it easily.

---

Cost Estimate

With Sonnet as the default and typical small-office usage:

• ~50 messages/day across all agents
• Roughly $5-15/month in API costs
• Research agent with Opus: budget extra for deep-dive sessions

---

Taking It Further

• Add cron jobs for scheduled tasks (morning email summary, Friday expense report, daily social metrics)
• Enable agent-to-agent messaging so Ops can ask Research to look something up
• Add Slack or Discord bindings alongside Telegram for teams that prefer those platforms
• Sandboxing for the Books agent if handling sensitive financial data