t2000: Agent Safeguards

Purpose

Configure spending limits and safety guardrails for autonomous agent operation. Three controls: agent lock (kill switch), per-transaction limit, and daily send limit.

Safeguards are enforced on all channels: CLI, MCP, and the AI Gateway (Telegram/WebChat). The gateway's confirmation flow is an additional layer — all state-changing actions require explicit "yes" before execution.

Commands

t2000 config show                    # view all safeguard settings
t2000 config set maxPerTx 500        # max $500 per outbound transaction
t2000 config set maxDailySend 1000   # max $1000 outbound per day
t2000 lock                           # freeze ALL operations immediately
t2000 unlock                         # resume operations (requires PIN)

Controls

What counts as outbound

• t2000 send — transfers to other addresses
• t2000 pay — x402 API payments
• t2000 sentinel attack — sentinel bounties

What is NOT limited

Internal operations (save, withdraw, borrow, repay, exchange, rebalance) move funds within the agent's own wallet and protocol positions. They are not subject to send limits.

Output (config show)

  Agent Safeguards
  ─────────────────────────────────
  Locked:             No
  Per-transaction:    $500.00
  Daily send limit:   $1,000.00 ($350.00 used today)

Blocked output

  ✗ Blocked: amount $1,000.00 exceeds per-transaction limit ($500.00)
  ✗ Blocked: daily send limit reached ($1,000.00/$1,000.00 used today)
  ✗ Agent is locked. All operations frozen.

JSON mode

t2000 config show --json

{ "locked": false, "maxPerTx": 500, "maxDailySend": 1000, "dailyUsed": 350 }

SDK

const agent = await T2000.create({ pin });

// Check config
agent.enforcer.getConfig();       // { locked, maxPerTx, maxDailySend, dailyUsed, ... }
agent.enforcer.isConfigured();    // true if any limit is non-zero

// Set limits
agent.enforcer.set('maxPerTx', 500);
agent.enforcer.set('maxDailySend', 1000);

// Lock/unlock
agent.enforcer.lock();
agent.enforcer.unlock();

Important

• Safeguards are opt-in — defaults are permissive (0 = unlimited)
• MCP server requires non-zero limits before starting
• Lock freezes ALL operations (including internal) as a safety measure
• Daily counter resets at midnight UTC