Open-Source Agent Infrastructure

The Operating System
for AI Agents

Deploy, monitor, govern, and scale production AI agents with built-in cost controls, safety policies, real-time observability, and a complete API-first platform.

quickstart.py
from acp_sdk import AgentClient

client = AgentClient("https://api.ai-control-plane.in")

# Register an agent
agent = client.agents.create(
    name="support-bot",
    model="gpt-4o",
    tools=["knowledge-base", "ticket-system"],
    policy={"max_tokens_per_run": 8000}
)

# Execute with full observability
run = client.agents.run(agent.id, {
    "input": "Resolve ticket #4521"
})

The Problem with AI Agents Today

Teams ship agents fast, but production demands more than a prompt and an API key.

No Visibility

Agents run as black boxes. You can't see what they're doing, how many tokens they burn, or why they fail.

Runaway Costs

A single misconfigured agent loop can drain your API budget in minutes with no safeguard in place.

No Safety Rails

Agents call tools without guardrails. One bad tool call can delete data, send wrong emails, or break systems.

Zero Audit Trail

No record of what agents decided, which tools they called, or why — making compliance impossible.

Everything You Need for Production Agents

23 integrated modules covering the full agent lifecycle — from registration to runtime to retirement.

Agent Registry & Versioning

Register, version, and manage agents with full configuration snapshots. Roll back to any previous version instantly.

  • Immutable version history
  • Config diffing
  • Agent lifecycle management

Real-Time Observability

Stream every agent run step-by-step. See token usage, latency, tool calls, and decisions as they happen.

  • Live run streaming
  • Token & cost tracking
  • Step-level tracing

Safety Policies & Guardrails

Define per-agent policies that enforce token limits, restrict tools, require approvals, and block dangerous actions.

  • Tool-level restrictions
  • Human-in-the-loop gates
  • Content filtering

Budget & Cost Controls

Set daily, weekly, or monthly spend limits per agent or team. Auto-block or throttle when thresholds hit.

  • Per-agent spend caps
  • Usage alerts at 80/90/100%
  • Block or throttle on exceed

Agent Memory

Give agents persistent memory across runs with semantic search. Session, episodic, and long-term memory types.

  • Vector search (pgvector)
  • 4 memory types
  • Cross-run context

Schedules & Automation

Run agents on cron schedules or trigger them via webhooks. Monitor schedule health and execution history.

  • Cron expressions
  • Manual trigger support
  • Schedule status tracking

Tool Registry

Register external tools (APIs, functions, databases) that agents can call. Test tools in isolation before attaching.

  • REST/function tool types
  • Input schema validation
  • In-dashboard testing

Production-Grade Architecture

Five distinct layers, each independently scalable. Deployed via Docker Compose with 12 containers.

Layer 1
Client SDK / API Calls
Python SDKREST APIWebhooksDashboard UI
Layer 2
API Gateway (Go)
Rate LimitingAuth (API Keys)Request RoutingLoad Balancing
Layer 3
Core API (FastAPI)
Agent RegistryRuntime EnginePolicy EngineBudget ManagerTool RegistryMemory StoreSchedulerAudit Logger
Layer 4
Workers (Celery)
Runtime WorkerBudget WorkerScheduler Worker
Layer 5
Data Layer
TimescaleDB + pgvectorRedis CacheKafka Events
23
API Modules
12
Docker Containers
120+
REST Endpoints
4
Database Extensions

Built for Every Agent Type

From customer support to DevOps to research — ACP provides the infrastructure for any autonomous agent.

Customer Support Agent

Resolves tickets by searching knowledge bases, updating CRMs, and escalating complex issues — all with audit trails.

Knowledge BaseTicket SystemEmail

Financial Analyst Agent

Pulls market data, runs calculations, generates reports, and flags anomalies with strict budget controls.

Market APICalculatorReport Generator

DevOps Agent

Monitors infrastructure, auto-scales services, rolls back deployments, and pages on-call — with safety policies.

AWS SDKKubernetes APIPagerDuty

Research Agent

Searches academic papers, summarizes findings, maintains persistent memory across research sessions.

arXiv APISemantic ScholarNote Store

Production-Ready from Day One

Enterprise features built in, not bolted on. Every layer designed for reliability, security, and scale.

Governance

  • Per-agent safety policies with tool restrictions
  • Human-in-the-loop approval gates
  • Max token and cost limits per run
  • Content filtering and output validation

Observability

  • Real-time run streaming with step tracing
  • Token usage, latency, and cost analytics
  • Time-series dashboards with 24h/7d/30d ranges
  • Agent performance leaderboards

Security & Audit

  • Complete decision audit ledger
  • API key authentication with scoped access
  • Event sourcing via Kafka
  • Per-run audit trail with tool call logs

Runtime

  • Async execution via Celery workers
  • Run queuing and cancellation
  • Multi-model support (GPT-4o, Claude, Gemini)
  • Cron scheduling with health monitoring

A Dashboard Built for Operators

17 modules in a single-page dashboard. Monitor, manage, and govern all your agents from one place.

ai-control-plane.in/dashboard

Overview

Total Runs
12,847
+18%
Active Agents
24
+3
Success Rate
96.4%
+2.1%
Total Spend
$1,284
-12%

See Every Step Your Agent Takes

Full run tracing with token counts, latency, tool calls, and costs. No black boxes.

COMPLETEDRun #run_8f2a · support-bot · 2.1s
Run Started0ms
Agent: support-bot v3
Tool: KB Search120ms
Query: 'refund policy'
LLM Reasoning1.2s
GPT-4o · 847 tokens
Tool: Update Ticket340ms
Ticket #4521 → Resolved
Run Complete2.1s
Total: 2,140 tokens · $0.018
All policy checks passed · Budget remaining: $48.72 / $50.00

Why AI Control Plane?

The AI industry is moving from single-prompt chatbots to autonomous agents that take actions, call tools, and make decisions. But production agents need more than a good prompt — they need operational infrastructure.

AI Control Plane is the missing layer between your agent code and production reality. It handles the hard parts — cost controls, safety guardrails, observability, audit logging, scheduling, and memory — so you can focus on building intelligent agents, not plumbing.

Think of it as Kubernetes for AI agents: you define what your agent should do, and ACP handles how it runs safely and reliably at scale.

15,000+
Lines of Backend Code
120+
API Endpoints
30+
Database Tables

Security Built Into Every Layer

Defense in depth — from API authentication to network isolation to complete audit logging.

API Key Authentication

Scoped API keys with role-based access. All requests authenticated at the gateway layer.

Complete Audit Trail

Every agent decision, tool call, and state change logged with timestamps and context.

Policy Enforcement

Declarative safety policies block dangerous tool calls, enforce token limits, and require approvals.

Input Validation

All agent inputs and tool parameters validated against JSON schemas before execution.

Network Isolation

Docker-based network isolation. Gateway, API, workers, and databases on separate internal networks.

Event Sourcing

All state changes emitted as Kafka events. Replay, audit, and integrate with external SIEM systems.

Built With

Battle-tested open-source technologies, composed into a production-grade platform.

Backend

Python 3.11FastAPICeleryasyncpg

Gateway

Go 1.22Chi RouterRate Limiter

Frontend

Next.js 14TypeScriptTailwind CSS

Database

TimescaleDBPostgreSQL 16pgvectorpgcrypto

Infrastructure

Docker ComposeRedisKafkaZookeeper

AI Models

GPT-4oClaude 3.5Gemini ProLlama 3

Simple, Transparent Pricing

Start free with the full open-source platform. Scale with managed hosting when you need it.

Open Source

Freeforever

Self-host the full platform. No limits, no strings attached.

  • All 23 API modules
  • Unlimited agents & runs
  • Full dashboard access
  • Docker Compose deployment
  • Community support
  • MIT License
Get Started
Most Popular

Pro

$99/month

Managed hosting with premium support and advanced features.

  • Everything in Open Source
  • Managed cloud hosting
  • Auto-scaling workers
  • SSO & team management
  • Priority email support
  • 99.9% uptime SLA
Coming Soon

Enterprise

Custom

Dedicated infrastructure, compliance, and hands-on support.

  • Everything in Pro
  • Dedicated infrastructure
  • Custom integrations
  • SOC 2 compliance support
  • Dedicated account manager
  • On-premise deployment option
Contact Us

Ready to Control Your AI Agents?

Deploy the full platform in minutes with Docker Compose, or explore the live demo right now.