Agent
Budget
KILL SWITCH
Real-time token spend guardrail for autonomous AI agents · Stop runaway before it hits your bill
Budget Configuration
Max Tokens
100,000
Max Spend (USD)
$50.00
Max Iterations
25
Warn at
80%
of budget
Model (cost per 1M tokens)
claude-opus-4-8 — $3.00/1M
claude-sonnet-5 — $1.00/1M
claude-haiku-4-5 — $0.25/1M
gpt-4o — $5.00/1M
gpt-4o-mini — $2.00/1M
o3 — $15.00/1M
Scenario
∞ Infinite Retry Loop
🔍 Deep Research Spiral
🔧 Tool Call Explosion
Customer support agent stuck in an error retry loop. API returns 503. Agent retries indefinitely.
critical risk
Guardrail Status
Token Consumption
0 / 100K
USD Spend
$0.00 / $50.00
Iteration Count
0 / 25
Live Agent Run
Ready to run — configure guardrails and press Start
iter:
0
0
Tokens Used
$0.00
USD Spent
—
Tok/s
Start Agent Run
Manual Kill
Reset
KILL SWITCH TRIGGERED — Agent Terminated
Token budget exceeded.
—
Tokens consumed
—
Cost at kill
—
Estimated saved
—
Iterations completed
Incident Report
Run Summary
Agent task
—
Termination trigger
—
Root cause
—
Duration
—
Tokens at kill
—
Cost at kill
—
Projected cost (unguarded)
—
Prevention Config (apply to agent init)
// loading...
AgentBudget · Built by
RLASAF12
· Inspired by the "$47→$5,847 in 58 minutes" incident (DEV.to, 2026-07-06) ·
View on GitHub