Now in public beta — Join early

Your AI agents are running blind.
See everything.

Monitor costs per agent, per session. Replay every step. Protect with guardrails.Works with Python, n8n, Make, Zapier, and any tool that calls an LLM.

No credit card required · Free tier forever · Set up in 2 minutes

2.4M+
Events tracked
500+
Developers
99.9%
Uptime
<2min
Setup time

Works with everything

OpenAI
Anthropic
LangChain
LlamaIndex
CrewAI
AutoGen
n8n
Make
Zapier
Flowise
REST API
cURL

“AgentShield caught a prompt loop that was burning $200/day on our production agent. Set up in under 2 minutes, budget cap activated, issue resolved.”

M
Marcus T.
Senior ML Engineer · Beta user

Monitor

MODULE

Track every dollar, in real-time

Cost tracking per agent, smart alerts with AI diagnosis, forecast costs that accumulate without per-agent visibility, and get automatic model recommendations.

Replay

MODULE

Debug in seconds, not hours

Step-by-step visual timeline of every agent session. See inputs, outputs, costs, and errors — then share with your team in one click.

Protect

MODULE

Guardrails that actually guard

Configurable content rules, automatic PII redaction, budget caps with kill switch, and compliance-ready audit logs.

Everything your agents need to behave.

One API call. Three modules. Complete observability for AI agents in production.

CORE

Real-time Cost Tracking

See every API call, every token, every dollar — the moment it happens. Not at the end of the month.

AI

Smart Alerts

When something goes wrong, AI diagnoses why and suggests a fix. Not just 'threshold exceeded'.

SAVINGS

Cost Autopilot

Automatic model recommendations with estimated savings. 'Switch to gpt-4o-mini here — save $85/month.'

DEBUG

Session Replay

Visual step-by-step timeline of every agent run. Inputs, outputs, costs — all in one view.

SECURITY

Guardrails & PII

Block unwanted content, redact personal data automatically, kill switch when things go wrong.

FORECAST

Cost Forecast

Know which agent caused the cost spike before it accumulates. Per agent, per team, with confidence intervals.

Up and running in 2 minutes.

No infrastructure. No config files. No code changes to existing agents.

01

Install the SDK

One line. pip install agentshield. Works with OpenAI, Anthropic, Google, LangChain, CrewAI, and more.

$ pip install agentshield
02

Add one decorator

@shield(agent='my-agent') — that's it. Costs, sessions, and guardrails are now tracked.

@shield(agent='my-agent')
async def research(query):
...
03

See everything

Open your dashboard. Costs in real-time. Sessions you can replay. Violations you can prevent.

today's spend$12.47
sessions1,247
violations0

One API call.
Full observability.

Any language. Any tool. Python, JavaScript, cURL, n8n, Make — if it can send an HTTP POST, it works with AgentShield.

Zero-latency overhead on your agent calls
Works with any LLM provider or framework
Sessions, costs, and guardrails out of the box
1import agentshield as shield
2 
3@shield(agent="my-research-agent")
4async def research_agent(query: str) -> str:
5 # AgentShield tracks per-agent, per-session:
6 # ✓ Token usage & costs
7 # ✓ Session timeline for replay
8 # ✓ Guardrail violations & anomalies
9 
10 response = await openai.chat.completions.create(
11 model="gpt-4o",
12 messages=[{"role": "user", "content": query}]
13 )
14 return response.choices[0].message.content
15 
16# That's it. Open your dashboard.
17# app.agentshield.one

Start free. Scale with confidence.

No credit card required. Upgrade or downgrade any time.

Free

For side projects

0forever

Get started with basic monitoring.

  • 1 agent
  • 10K requests/month
  • 7-day history
  • Monitor (basic)
  • Community support

Starter

For growing teams

49/month

Add replay and smart alerts.

  • 5 agents
  • 100K requests/month
  • 30-day history
  • Monitor + Replay
  • Alerts & Anomaly detection
  • Forecast
  • Email support
POPULAR

Pro

For production agents

99/month

Full suite: Monitor + Replay + Protect.

  • Unlimited agents
  • 500K requests/month
  • 90-day history
  • Monitor + Replay + Protect
  • Smart Alerts (AI diagnosis)
  • Cost Autopilot
  • Webhooks & Session sharing
  • Budget caps & kill switch
  • Priority support

Team

For agencies & enterprises

199/month

Multi-user, compliance, and everything else.

  • Everything in Pro
  • Multi-user & Teams
  • Team cost attribution
  • Audit log (immutable)
  • PDF reports
  • Slack bot
  • 1-year history
  • Dedicated support

Why AgentShield?

Built specifically for AI agents — not retrofitted from generic APM tools.

FeatureYOU ARE HEREAgentShieldLangSmithHeliconeLangfuseProvider Limits
Real-time cost tracking✅ total only
Per-agent cost breakdown
Per-session cost breakdown
Session Replay
Guardrails & PII redaction
Budget caps per agent✅ total only
Kill switch per agent
Anomaly detection
No-code support (n8n, Make, Zapier)
REST API (no SDK required)
Free tier

OpenAI and Anthropic give you total spending limits. AgentShield gives you per-agent, per-session cost attribution. Your support-agent has a $5/day budget. Your research-agent has $50/day. Provider dashboards can't do that.

“But I already have spending limits on OpenAI”

Provider limits cap your total. AgentShield shows you exactly where.

What provider limits show you

Total spend: $340 this month.
Rate limit: 10,000 req/min.
Credit cap: $500/month.

That's it.

What AgentShield shows you

support-agent:   $12/month (normal)
research-agent:  $280/month (⚠ 3x baseline)
  → session #4847 looped 97 times
  → cost: $40 in one session
  → anomaly detected → Slack alert
  → budget cap hit → agent frozen

Know WHERE, not just how much.

Frequently asked questions

No. AgentShield tracks events asynchronously — your agent calls return immediately, with zero added latency. Events are batched and sent in the background.

OpenAI, Anthropic, Google (Gemini), Cohere, and any OpenAI-compatible API. Framework support: LangChain, CrewAI, AutoGen, LlamaIndex, and plain Python.

The SDK detects and redacts email addresses, phone numbers, credit card numbers, SSNs, and IP addresses before data leaves your environment. You can configure custom patterns.

Yes. Outbound webhooks (Pro+) let you forward events to Datadog, Grafana, PagerDuty, or any HTTP endpoint. AgentShield complements — not replaces — your existing tools.

Event data is stored per your plan: 7 days (Free), 30 days (Starter), 90 days (Pro), 1 year (Team). Session inputs/outputs are only stored if you opt in, and always with PII redacted.

AgentShield raises a BudgetExceededError in your code and optionally activates the kill switch — preventing further API calls until you lift it manually or the period resets.

Not yet. We are working on a self-hosted version for enterprise compliance needs. Contact us if that is a requirement.

No. AgentShield has a REST API that works with any tool that can send an HTTP POST request — n8n, Make, Zapier, Retool, any programming language. The Python SDK is optional.

Add an HTTP Request node after your AI call. POST to api.agentshield.one/v1/track with your agent name and token counts. Costs appear in the dashboard in real time. No SDK, no code changes.

AgentShield is the only tool that combines per-agent cost attribution, session replay, guardrails, and budget kill switches in one product. It also works with no-code tools like n8n and Make, which none of the alternatives support.

Provider limits cap your total spend. AgentShield caps per agent, per session. Your OpenAI dashboard shows 'you spent $340 this month.' AgentShield shows 'your research-agent spent $280, and session #4847 looped 97 times causing a $40 spike on Tuesday.' That per-agent attribution is what providers don't give you.

No. Providers show total spend. AgentShield tracks per-agent, per-session costs, replays every step of an agent session, detects anomalies using statistical baselines, enforces budget caps per agent with automatic kill switches, and redacts PII from tracked data. None of that exists in any provider dashboard.

Your agents are running.

Do you know what they're doing?

Set up in 2 minutes. No credit card. Free tier forever.