See What Your AI Agents Actually Do

Runtime visibility, audit trails, and kill switches for LangChain, CrewAI, AutoGen, and custom agents. Open source. Local-first. No data leaves your machine.

Get Started Free
Star Open Source
Package PyPI Package
License Apache 2.0
Plugin MCP Compatible

The Problem

AI agents are sending emails, updating CRMs, and making decisions. Nobody knows what they're actually doing.

Blind Spots

Agents run autonomously but logs are useless to non-engineers. What did it actually do? You'll need to dig through a mountain of debug output.

No Kill Switch

Can't stop a rogue agent without killing the whole system. If an agent goes haywire, your only option is a hard shutdown.

Audit Failures

EU AI Act Article 12 requires tamper-evident records. Logs in a database? Good luck proving nothing was modified after the fact.

3 Lines. Any Framework.

Works with LangChain, CrewAI, AutoGen, custom agents - whatever you're building.

from air_controls import setup_langchain_controls from langchain.agents import create_openai_functions_agent air = setup_langchain_controls() agent = create_openai_functions_agent(llm, tools, prompt) air.hook_agent(agent)
from air_controls import audit_action @audit_action(agent_name="my-bot") def process_request(request): return llm.complete(request) # Every call to process_request is now logged & audited result = process_request("Do something important")
from air_controls import setup_crewai_controls air = setup_crewai_controls() crew = Crew(agents=[agent1, agent2], tasks=[task1, task2]) air.hook_crew(crew) crew.kickoff()
from air_controls.mcp import MCPServer server = MCPServer(agent_name="research-bot") server.add_tool(search_web) server.add_tool(summarize) server.run() # Works with Cursor, Claude Code, Windsurf

Install from PyPI:

pip install air-controls

Interactive Dashboard Demo

Real-time event timeline and audit chain verification

Agent Monitor
0
Total Events
3
Active Agents
Valid
Chain Status
100%
Uptime

Built for Production

Chart

Action Timeline

Every LLM call, tool use, and API request logged with full context and timing. See exactly what your agents are doing in real time.

Stop

Kill Switch

Pause any agent instantly from CLI or dashboard. No more runaway agents. Stop, investigate, resume safely.

Lock

HMAC-SHA256 Audit Chain

Tamper-evident records for compliance audits. Each event cryptographically linked to the previous one. Auditors can verify nothing was modified.

Plug

Framework Agnostic

LangChain, CrewAI, AutoGen, OpenAI, custom agents-bring your own. Single integration point, works everywhere.

Tools

MCP Server

Works with Cursor, Claude Code, Windsurf automatically. Drop it into your AI workflow. IDE integration out of the box.

Home

Local-First

SQLite. No cloud. No data leaves your machine. Full control over your audit logs. Run it on your laptop or production infra.

Simple Pricing

Pay for what you use. No lock-in.

Free
$0
/month
  • Unlimited agents
  • CLI dashboard
  • Event logging
  • HMAC audit chain
  • Open source
Get Started
Enterprise
$99
/agent/month
  • SSO / SAML
  • SLA guarantee
  • Custom integrations
  • Dedicated support
  • Unlimited team seats
  • Compliance reports
Contact Sales
Volume discounts available

How It All Fits Together

AIR Controls is the runtime counterpart to AIR Blackbox

AIR Blackbox
Scans code pre-deploy for EU AI Act compliance issues (Articles 9-15)
Down arrow
AIR Controls
Monitors what agents actually do at runtime with tamper-evident logs
Down arrow
Same HMAC-SHA256 audit chain for end-to-end compliance