AI Agents

Agent Administration

Manage your AI agents through messaging commands. Control approval modes, trust tiers, confidence thresholds, and policies directly from Telegram or WhatsApp.

Messaging Commands

All commands can be used in your connected messaging channel (Telegram or WhatsApp). The bot will respond with confirmation and current state.

/agents

List all agents with their current status and trust tier.

/agents

// Response:
// 🤖 Active Agents:
// ┌──────────────────┬───────────┬────────────┬──────────┐
// │ Agent            │ Status    │ Trust Tier  │ Mode     │
// ├──────────────────┼───────────┼────────────┼──────────┤
// │ pricing_agent    │ ✅ Active  │ Supervised  │ auto     │
// │ deal_agent       │ ✅ Active  │ Trusted     │ auto     │
// │ inventory_agent  │ ⏸️ Paused  │ Probation   │ approve  │
// │ content_agent    │ 🚫 Off    │ Probation   │ disabled │
// └──────────────────┴───────────┴────────────┴──────────┘
/agentmode <agent> <mode>

Set the approval mode for an agent. Modes: auto, approve, notify, disabled.

/agentmode pricing_agent auto

// Response:
// ✅ pricing_agent mode set to auto.
// Actions will be auto-approved or held based on trust tier.

/agentmode deal_agent approve

// Response:
// ✅ deal_agent mode set to approve.
// All actions will require manual approval.
/mintier <agent> <tier>

Set the minimum trust tier for an agent. Tiers: probation, supervised, trusted, autonomous.

/mintier pricing_agent supervised

// Response:
// ✅ Minimum tier for pricing_agent set to Supervised.
// Agent will not drop below this tier.
/threshold <agent> <value>

Set the confidence threshold for an agent (0.0 to 1.0).

/threshold pricing_agent 0.85

// Response:
// ✅ Confidence threshold for pricing_agent set to 0.85.
// Actions with confidence below 0.85 will be escalated.
/policy

View current policies for all agents.

/policy

// Response:
// 📋 Current Agent Policies:
//
// pricing_agent:
//   approval_mode: auto
//   confidence_threshold: 0.85
//   max_actions_per_hour: 50
//   max_financial_impact: $1,000
//   trust_tier: supervised
//   min_tier: supervised
//
// deal_agent:
//   approval_mode: auto
//   confidence_threshold: 0.7
//   max_actions_per_hour: 30
//   max_financial_impact: $500
//   trust_tier: trusted
//   min_tier: probation
/policy set <key> <value>

Update a specific policy value.

/policy set pricing_agent.max_actions_per_hour 25

// Response:
// ✅ pricing_agent.max_actions_per_hour set to 25.

/policy set deal_agent.max_financial_impact 750

// Response:
// ✅ deal_agent.max_financial_impact set to $750.

MerchantAgentPolicy Model

The MerchantAgentPolicy model stores all agent configuration for a merchant. It includes global policies derived from your merchant tier and per-agent overrides.

// MerchantAgentPolicy model
{
  "merchant_id": "merch_xyz789",
  "merchant_tier": "growth",        // newcomer/starter/growth/established/legendary
  "global_policies": {
    "max_discount_pct": 25,         // Maximum discount any agent can create
    "daily_transaction_limit": 10000,
    "max_agents": 4
  },
  "agent_policies": {
    "pricing_agent": {
      "approval_mode": "auto",
      "confidence_threshold": 0.85,
      "max_actions_per_hour": 50,
      "max_financial_impact": 1000,
      "trust_tier": "supervised",
      "min_tier": "supervised",
      "trust_tier_auto_upgrade": true
    },
    "deal_agent": {
      "approval_mode": "auto",
      "confidence_threshold": 0.7,
      "max_actions_per_hour": 30,
      "max_financial_impact": 500,
      "trust_tier": "trusted",
      "min_tier": "probation",
      "trust_tier_auto_upgrade": true
    }
  }
}

Tier-Based Policies

Your merchant tier determines the upper bounds for agent capabilities. These limits cannot be exceeded regardless of individual agent guardrail settings.

TierRequirementsMax DiscountDaily LimitAgents
NewcomerNew account10%$5001
Starter7+ days, 10+ orders15%$2,0002
Growth30+ days, 50+ orders25%$10,0004
Established90+ days, 200+ orders35%$50,0006
Legendary180+ days, 1000+ orders50%$200,00010

Emergency Disable

If an agent is behaving unexpectedly, you can immediately disable it. This cancels all pending actions and prevents any new actions until you re-enable the agent.

// Emergency disable - immediately stops all agent actions
/agentmode pricing_agent disabled

// Response:
// 🚫 pricing_agent has been DISABLED.
// All pending actions have been cancelled.
// No new actions will be taken until re-enabled.
//
// To re-enable: /agentmode pricing_agent auto

// You can also disable all agents at once:
/agentmode all disabled

// Response:
// 🚫 All agents have been DISABLED.
// 4 pending actions cancelled.
// Use /agentmode <agent> <mode> to re-enable individually.

What's Next?