agentguard-ram 0.1.0

Real-time cost observability and guardrails for AI agents

AgentGuard

Real-time cost observability and guardrails for AI agents.

Track LLM spend, set budget thresholds, and auto-pause agents before they blow up your bill.

Install

pip install agentguard

Quick Start

from agentguard import AgentGuard, AgentEvent

guard = AgentGuard(
    api_key="ag_...",
    agent_id="my-sales-agent",
    cost_threshold=10.0,          # auto-pause at $10/day
    slack_webhook="https://...",  # optional alert
)

# Custom agent — wrap with decorator
@guard.track
def run_my_agent():
    # your agent logic here
    alive = guard.send_event(AgentEvent(
        agent_id="my-sales-agent",
        event_type="llm_call",
        model="gpt-4o",
        input_tokens=500,
        output_tokens=300,
    ))
    if not alive:
        print("Agent paused — cost threshold reached")
        return

LangChain / LangGraph

from agentguard import AgentGuard

guard = AgentGuard(api_key="ag_...", agent_id="my-agent", cost_threshold=5.0)

# Drop-in callback — tracks every LLM call automatically
chain = my_chain.with_config(callbacks=[guard.callback])
result = chain.invoke({"input": "do something"})

Supported Models

Model	Input (per 1M)	Output (per 1M)
gpt-4o	$2.50	$10.00
gpt-4o-mini	$0.15	$0.60
claude-3-5-sonnet	$3.00	$15.00
claude-3-5-haiku	$0.80	$4.00
gemini-1.5-pro	$1.25	$5.00

Dashboard

View live costs, event feed, and manage agents at agent-guard-nine.vercel.app.

Links

• Dashboard: https://agent-guard-nine.vercel.app
• Backend API: https://agent-guard-production-dcdd.up.railway.app/docs