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.
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.
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.
Financial Analyst Agent
Pulls market data, runs calculations, generates reports, and flags anomalies with strict budget controls.
DevOps Agent
Monitors infrastructure, auto-scales services, rolls back deployments, and pages on-call — with safety policies.
Research Agent
Searches academic papers, summarizes findings, maintains persistent memory across research sessions.
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.
Overview
See Every Step Your Agent Takes
Full run tracing with token counts, latency, tool calls, and costs. No black boxes.
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.
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
Gateway
Frontend
Database
Infrastructure
AI Models
Simple, Transparent Pricing
Start free with the full open-source platform. Scale with managed hosting when you need it.
Open Source
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
Pro
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
Enterprise
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
Ready to Control Your AI Agents?
Deploy the full platform in minutes with Docker Compose, or explore the live demo right now.