AI Agent Governance Platform

You deployed 30 agents but can't answer basic questions about any of them. Which ones are running right now? How much have they spent today? Has any of them accessed a tool it shouldn't have? AXME is a governance platform for production AI agents.

Alpha - Built with AXME (AXP Intent Protocol). cloud.axme.ai - contact@axme.ai

The Problem

You have 30 AI agents in production. Right now, you cannot answer any of these questions:

1. Which agents are running right now?
   - No heartbeat monitoring. No health checks. You check logs manually.

2. How much has each agent spent today?
   - No cost tracking. You find out at the end of the month from the OpenAI bill.

3. Has any agent used a tool it shouldn't have?
   - No tool permissions. Every agent can call every tool.

4. Can you shut down a specific agent immediately?
   - No kill switch. You SSH into the server and kill the process.

5. What did each agent do in the last 24 hours?
   - No audit trail. Actions are scattered across stdout logs on 12 machines.

Every team hits this wall around agent #10. The agents work individually, but nobody governs them as a fleet.

What breaks without governance:

No visibility - agents run in the dark. You learn about failures from customers.
No cost control - one runaway agent can burn your entire monthly LLM budget overnight.
No permissions - every agent has access to every tool. A report-writing agent can delete the database.
No kill switch - when something goes wrong, you scramble to find which process on which server to kill.
No audit trail - when the CEO asks "what did the agents do yesterday?", you have no answer.

The Solution: AXME Agent Mesh

AXME provides a governance layer that sits between your agents and the outside world. Every agent registers, reports health, obeys policies, and logs actions to a central audit trail.

                          AXME Agent Mesh
                    +--------------------------+
                    |                          |
 Register  ------>  |  Agent Registry          |
                    |  - fleet inventory       |
 Heartbeat ------>  |  - health monitoring     |
                    |  - cost tracking         |
 Policy    ------>  |  - tool permissions      |
                    |  - approval thresholds   |
 Kill      ------>  |  - kill switch           |
                    |  - audit trail           |
 Dashboard ------>  |  - real-time dashboard   |
                    |                          |
                    +--------------------------+
                              |
              Delivers governance intents
              to all agents via AXP protocol

What you get

Feature	What it does
Agent Registry	Inventory of all agents with metadata (team, framework, model, environment)
Health Monitoring	Heartbeat-based health checks. Agents report status, latency, memory, CPU
Cost Tracking	Real-time per-agent cost with configurable caps and warnings at 80%
Tool Permissions	Per-agent allowlist of tools. Unauthorized tool calls are blocked and logged
Approval Gates	Tasks above a cost threshold require human approval before execution
Kill Switch	Shut down a single agent or the entire fleet instantly with one command
Audit Trail	Every governance event logged: registrations, violations, kills, approvals
Dashboard	Real-time fleet overview with health, cost, latency, and violation status

Quick Start

Prerequisites

# Install CLI
curl -fsSL https://raw.githubusercontent.com/AxmeAI/axme-cli/main/install.sh | sh

# Log in
axme login

# Install Python SDK
pip install axme

Step 1: Register your agent fleet

export AXME_API_KEY="your-key"
python register_agents.py

Registering agent fleet...

  Registered: Data Analyst Agent
    Address:  gov-data-analyst
    Cost cap: $50.0
    Tools:    ['sql_query', 'chart_generate', 'export_csv']

  Registered: Report Generator Agent
    Address:  gov-report-generator
    Cost cap: $30.0
    Tools:    ['read_file', 'write_report', 'send_email']

  Registered: Code Reviewer Agent
    Address:  gov-code-reviewer
    Cost cap: $100.0
    Tools:    ['read_repo', 'post_comment', 'approve_pr']

Fleet registered: 3 agents

Step 2: Start heartbeat monitoring

# In separate terminals, one per agent:
AXME_API_KEY=<agent-key> python heartbeat_agent.py --agent gov-data-analyst
AXME_API_KEY=<agent-key> python heartbeat_agent.py --agent gov-report-generator
AXME_API_KEY=<agent-key> python heartbeat_agent.py --agent gov-code-reviewer

Heartbeat reporter for gov-data-analyst
Interval: 30s
Sending heartbeats (Ctrl+C to stop)

  [1] gov-data-analyst | cost=$0.42 | latency=1150ms | requests=3 | intent=axme_82f1a3...
  [2] gov-data-analyst | cost=$0.85 | latency=1340ms | requests=7 | intent=axme_91c2b4...

Step 3: Start the policy enforcer

AXME_API_KEY=<agent-key> python policy_enforcer.py

Policy enforcer running...
Monitoring 3 agents
Listening for heartbeats and tool requests (Ctrl+C to stop)

  ** VIOLATION: cost_cap_warning for gov-report-generator **
     Cost: $24.50 / Cap: $30.00

  ** BLOCKED: gov-data-analyst tried to use unauthorized tool 'delete_table' **

Step 4: View the fleet dashboard

AXME_API_KEY="your-key" python dashboard.py

==========================================================================================
  AI AGENT GOVERNANCE DASHBOARD
  2026-03-31 14:22:08 UTC
==========================================================================================

  Agent                     Health       Cost            Latency      Requests   Policy
  ------------------------- ------------ --------------- ------------ ---------- ----------
  gov-data-analyst          HEALTHY      $12.50/$50      1200ms       45         OK
  gov-report-generator      HEALTHY      $24.50/$30      2400ms       22         WARNING
  gov-code-reviewer         HEALTHY      $38.20/$100     3100ms       15         OK

------------------------------------------------------------------------------------------
  Fleet: 3/3 healthy | Total cost: $75.20/$180 | Budget used: 42%

  ACTIVE VIOLATIONS:
    [COST]   gov-report-generator: $24.50 approaching cap $30

==========================================================================================

Step 5: Kill switch (when needed)

# Kill a single agent
python kill_switch.py --agent gov-data-analyst --reason "cost cap exceeded"

# Kill the entire fleet
python kill_switch.py --all --reason "security incident"

KILL SWITCH ACTIVATED
  Reason:   security incident
  Operator: admin
  Targets:  3 agent(s)

  KILLED: gov-data-analyst (intent: axme_f3a1...)
  KILLED: gov-report-generator (intent: axme_b2c4...)
  KILLED: gov-code-reviewer (intent: axme_d1e5...)

  Audit log: axme_a9f2...

3 agent(s) killed. Check dashboard for status.

Step 6: View audit trail

AXME_API_KEY=<agent-key> python audit_trail.py

Governance Audit Trail
======================================================================
Listening for audit events (Ctrl+C to stop)

--- Event #1 ---
  [2026-03-31T14:20:12Z] cost_warning
    Agent:  gov-report-generator
    Source: policy-enforcer
    Cost:   $24.50 / $30.00

--- Event #2 ---
  [2026-03-31T14:21:45Z] tool_blocked
    Agent:  gov-data-analyst
    Source: policy-enforcer
    Tool:   delete_table
    Allowed: ['sql_query', 'chart_generate', 'export_csv']

--- Event #3 ---
  [2026-03-31T14:22:08Z] kill_switch_activated
    Agent:  ['gov-data-analyst', 'gov-report-generator', 'gov-code-reviewer']
    Source: unknown
    Reason:   security incident
    Operator: admin

Before / After

Before: DIY governance (1000+ lines, fragile)

# Health monitoring: custom cron job checking each agent
# Cost tracking: parse OpenAI invoices monthly
# Tool permissions: hope nobody calls the wrong function
# Kill switch: SSH into server, find PID, kill -9
# Audit trail: grep through CloudWatch logs across 12 services
# Dashboard: spreadsheet updated weekly by hand

After: AXME governance (6 files, production-ready)

from axme import AxmeClient, AxmeClientConfig

client = AxmeClient(AxmeClientConfig(api_key=os.environ["AXME_API_KEY"]))

# Register agent with policies
client.send_intent({
    "intent_type": "intent.governance.register_agent.v1",
    "to_agent": "agent://myorg/production/data-analyst",
    "payload": {
        "agent_address": "data-analyst",
        "policies": {
            "cost_cap_usd": 50.0,
            "allowed_tools": ["sql_query", "chart_generate"],
            "require_approval_above_usd": 25.0,
        },
    },
})

# Send heartbeat
client.send_intent({
    "intent_type": "intent.governance.heartbeat.v1",
    "to_agent": "agent://myorg/governance/monitor",
    "payload": {"agent_address": "data-analyst", "status": "healthy", "metrics": {...}},
})

# Kill switch
client.send_intent({
    "intent_type": "intent.governance.kill_switch.v1",
    "to_agent": "agent://myorg/production/data-analyst",
    "payload": {"action": "kill", "reason": "cost cap exceeded"},
})

No cron jobs. No SSH. No log parsing. No spreadsheets.

How It Works

+----------+     register      +------------------+
| Agent 1  | --------------->  |                  |
| Agent 2  | -- heartbeat -->  |   AXME Cloud     |     +-------------+
| Agent 3  | <-- policies ---  |   (governance)   | --> | Audit Trail |
|   ...    | <-- kill -------  |                  |     +-------------+
| Agent 30 |                   |   intent state   |
+----------+                   |   in PostgreSQL  |     +-------------+
                               |   (durable)      | --> | Dashboard   |
+----------+                   |                  |     +-------------+
| Policy   | -- enforce ---->  |                  |
| Enforcer | <-- violations -  |                  |
+----------+                   +------------------+

Register agents with metadata and policies via governance intents
Heartbeat - each agent reports health, cost, latency every 30 seconds
Policy enforcer listens to heartbeats, checks cost caps and tool permissions
Violations trigger alerts, blocks, or kill switch automatically
Kill switch sends a durable intent the agent must honor
Audit trail logs every governance event to a persistent store
Dashboard shows real-time fleet status with violation highlights

All governance actions are durable AXP intents. If the policy enforcer crashes, it restarts and picks up where it left off. If an agent crashes, the kill intent waits and delivers when it comes back.

Works With Any Agent Framework

AXME governance wraps around your existing agents. No framework changes.

Framework	How It Works
LangGraph	Add heartbeat + policy check to your graph nodes
CrewAI	Wrap crew tasks with governance intents
AutoGen	Add governance layer to agent conversations
OpenAI Agents SDK	Intercept tool calls through policy enforcer
Google ADK	Same pattern - register, heartbeat, obey policies
Raw Python	`send_intent()` for every governance action

Run the Full Example

Prerequisites

curl -fsSL https://raw.githubusercontent.com/AxmeAI/axme-cli/main/install.sh | sh
axme login
pip install axme

Deploy the scenario

axme scenarios apply scenario.json

Get agent keys

# macOS
cat ~/Library/Application\ Support/axme/scenario-agents.json

# Linux
cat ~/.config/axme/scenario-agents.json

Run all components

# Terminal 1: Register fleet
AXME_API_KEY="your-key" python register_agents.py

# Terminal 2: Policy enforcer
AXME_API_KEY=<enforcer-key> python policy_enforcer.py

# Terminal 3: Heartbeat (one per agent)
AXME_API_KEY=<agent-key> python heartbeat_agent.py --agent gov-data-analyst

# Terminal 4: Dashboard
AXME_API_KEY="your-key" python dashboard.py

# Terminal 5: Audit trail
AXME_API_KEY=<audit-key> python audit_trail.py

# When needed: Kill switch
AXME_API_KEY="your-key" python kill_switch.py --agent gov-data-analyst --reason "testing"

AXME - project overview
AXP Spec - open Intent Protocol specification
AXME Examples - 20+ runnable examples across 5 languages
AXME CLI - manage intents, agents, scenarios from the terminal
AI Agent Health Monitoring - heartbeat-based health checks
AI Agent Kill Switch - emergency shutdown for agents
AI Agent Cost Monitoring - per-agent cost tracking and caps
AI Agent Fleet Dashboard - real-time fleet visibility

Built with AXME (AXP Intent Protocol).

Navigation Menu

Uh oh!

Repository files navigation

AI Agent Governance Platform

The Problem

The Solution: AXME Agent Mesh

What you get

Quick Start

Prerequisites

Step 1: Register your agent fleet

Step 2: Start heartbeat monitoring

Step 3: Start the policy enforcer

Step 4: View the fleet dashboard

Step 5: Kill switch (when needed)

Step 6: View audit trail

Before / After

Before: DIY governance (1000+ lines, fragile)

After: AXME governance (6 files, production-ready)

How It Works

Works With Any Agent Framework

Run the Full Example

Prerequisites

Deploy the scenario

Get agent keys

Run all components

Related

About

Releases

Packages

Contributors

Languages

Name		Name	Last commit message	Last commit date
LICENSE		LICENSE
README.md		README.md
audit_trail.py		audit_trail.py
dashboard.py		dashboard.py
heartbeat_agent.py		heartbeat_agent.py
kill_switch.py		kill_switch.py
mesh-dashboard.png		mesh-dashboard.png
mesh-policies.png		mesh-policies.png
policy_enforcer.py		policy_enforcer.py
register_agents.py		register_agents.py
requirements.txt		requirements.txt
scenario.json		scenario.json

Uh oh!

Latest commit

History

Folders and files

Repository files navigation

AI Agent Governance Platform

The Problem

The Solution: AXME Agent Mesh

What you get

Quick Start

Prerequisites

Step 1: Register your agent fleet

Step 2: Start heartbeat monitoring

Step 3: Start the policy enforcer

Step 4: View the fleet dashboard

Step 5: Kill switch (when needed)

Step 6: View audit trail

Before / After

Before: DIY governance (1000+ lines, fragile)

After: AXME governance (6 files, production-ready)

How It Works

Works With Any Agent Framework

Run the Full Example

Prerequisites

Deploy the scenario

Get agent keys

Run all components

Related

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages