← derez.ai Home

What Is an AI Agent?

An AI agent is a software program that uses a large language model (LLM) as its reasoning engine and has the ability to act on its environment — browsing the web, reading and writing files, sending emails, calling APIs, and executing code. Unlike a standard chatbot that only responds to prompts, an agent can perform multi-step tasks autonomously.

Key characteristics that define an AI agent:

• LLM-powered reasoning — connects to models like GPT-4, Claude, DeepSeek, or open-source alternatives to interpret tasks and plan actions
• Tool access — can browse the web, read files, run code, send emails, connect to external APIs and databases
• Autonomous execution — given a goal, the agent determines the steps, executes them, and reports results without step-by-step human guidance
• Persistent state — has memory, skills, and configuration that persist across conversations and tasks

The Two Parts of an AI Agent System

A common misconception is that an AI agent is a single piece of software that runs on one machine. In reality, every production AI agent system consists of two separate components that are best kept on separate hardware:

Part 1: The Agent Orchestrator

This is the software that manages the agent's reasoning loop — interpreting user input, deciding what tools to call, executing actions, and maintaining conversation state. Examples include Hermes Agent, OpenClaw, Pi Agent, LangChain, AutoGPT, and CrewAI. The orchestrator runs on a standard Linux server with:

• A CPU (no GPU required — it makes API calls to the LLM, it doesn't run one)
• RAM for the runtime and tool execution (typically 1–4 GB)
• Storage for skills, plugins, cron jobs, logs, and cached data (typically 5–10 GB)
• Network access to the LLM provider API and any integration endpoints

The agent orchestrator is lightweight. It can run on a $5–15/month VPS comfortably.

Part 2: The Inference Provider

This is where the actual LLM runs — the model that processes prompts and generates responses. Models like GPT-4, Claude, DeepSeek, Llama, or Mistral. Inference can be provided in two ways:

• API-based (recommended): Rent inference from a provider like OpenRouter, OpenAI, or Anthropic. Pay per token — typically $0.15–$2 per million tokens depending on the model. No GPU hardware to manage, no capacity planning, no model updates.
• Self-hosted: Run an open-weight model on your own GPU hardware. Requires a powerful GPU (24 GB+ VRAM for decent models), power, cooling, and maintenance. The cost of the GPU alone ($3,000–$30,000) dwarfs the agent's server cost.

Why Keep Them Separate?

Running the agent orchestrator and the inference engine on the same machine is inefficient for several reasons:

• Resource mismatch: The agent orchestrator needs CPU, RAM, and storage. The inference engine needs VRAM and GPU compute. These don't overlap — one is always idle while the other works.
• Wasted GPU hours: A GPU that's also running the agent orchestrator sits mostly idle while the agent processes tools, reads files, or waits for API responses. GPUs are the most expensive component — you want them running inference 100% of the time, not waiting for tool calls.
• Independent scaling: As you add more skill integrations (Odoo, email, web research), the agent needs more CPU and storage — but the same inference capacity. As your traffic grows, you need more inference throughput — but the same agent capacity. Separate hardware lets you scale each independently.
• Cost efficiency: API-based inference costs pennies per task. A GPU server costs $100–$1,000+/month. Pairing a $10/month agent VPS with pay-per-token inference is dramatically cheaper than running a GPU server just to host one agent's LLM calls.

How AI Agents Connect to LLMs

An AI agent does not run an LLM locally. Instead, it connects to an LLM provider through an API. This is a critical architectural distinction that affects hosting requirements:

• API-based access — The agent sends structured prompts to an LLM provider (OpenAI, Anthropic, DeepSeek, etc.) and receives generated responses. The agent handles the orchestration — deciding what to ask the LLM, executing tool calls, and chaining multiple LLM invocations together.
• Provider flexibility — Multiple LLMs can be configured and swapped without changing the agent's code. The same agent can use GPT-4 for reasoning tasks, Claude for long-context analysis, and DeepSeek for cost-sensitive operations.
• No local GPU required — Because the LLM runs on the provider's infrastructure, the agent's server does not need expensive GPU hardware. This is why agent hosting is fundamentally different from model hosting.

AI Agent Hosting Requirements

Hosting an AI agent requires more than a simple web server. An agent is a full Linux system with runtime dependencies, tool integrations, and persistent state:

Runtime Environment

• Python or Node.js runtime for the agent framework (e.g., Hermes Agent, LangChain, AutoGPT)
• Package managers (pip, npm, uv) for installing skills and dependencies
• System packages for tool integrations (image processing, document parsing, database connectors)

Tool Integrations

• Web browser / search API for research tasks
• Email (SMTP/IMAP) for sending and receiving messages
• File storage integrations (Nextcloud, S3, local filesystem)
• Database and ERP connectors (Odoo, PostgreSQL, etc.)
• API keys for LLM providers and third-party services

State and Persistence

• Conversation memory and task history
• Installed skills, plugins, and custom instructions
• Cron jobs, scheduled workflows, and automation rules
• Model caches, package caches, and downloaded dependencies

Popular AI Agent Frameworks

The following agent frameworks are actively maintained and production-ready as of 2026:

Hermes Agent (by Nous Research)

An open-source agent framework with a built-in skill system, tool library, and cron scheduler. Hermes runs as a non-root user with dedicated tool-access permissions. It supports Telegram, Discord, and web UI as conversation interfaces. The skill system allows adding domain-specific capabilities (cold email, Odoo integration, competitor profiling) without modifying core code. Official Documentation →

Other Frameworks

LangChain, AutoGPT, CrewAI, and OpenAI Assistants API offer alternative approaches to agent design. Most share the same architectural pattern: LLM API access, tool execution, and persistent state management.

What to Look for in an AI Agent Hosting Provider

Based on real operational experience with production agent deployments, here are the criteria that matter:

1. Full Linux instance, not container snapshots — Agents need a real operating system with package managers, cron, SSH access, and persistent storage. Lightweight containers that reset on restart are incompatible with agent workloads.
2. Full-disk backups — As shown in the backup analysis, agents scatter state across the entire filesystem. A proper backup captures everything, not just the home directory.
3. SSH access — The ability to inspect logs, install packages, debug issues, and configure low-level system settings is essential for maintaining production agents.
4. Managed updates — The agent framework, runtime, and dependencies should be maintained by the hosting provider. The user manages the agent's prompt and skills, not its operating system.
5. API key management — LLM provider keys, integration credentials, and tool API tokens must be stored securely and configurable through a dashboard.

How derez.ai Implements These Standards

derez.ai is a managed AI agent hosting platform built on these principles:

• Each agent runs on its own dedicated Linux instance with full root SSH access
• Automatic full-disk daily backups with 7-day retention and one-click restore
• Pre-installed Hermes Agent framework with access to a growing library of skills
• Dashboard for managing agents, passwords, API keys, integrations, and backups
• Coupon codes reduce the first month to as low as $0.00 for evaluation

Further Reading

• Hermes Agent Setup Guide — From zero to a running agent
• Why Backing Up an AI Agent Is So Hard — Real filesystem analysis
• Direct Odoo API vs MCP — Architecture comparison for ERP-connected agents
• Speed Run: Cold Email Agent — Building an email-sending agent
• Speed Run: Odoo Community Agent — Building an ERP-connected agent

derez.ai — Managed AI agent hosting. · Blog · Focus Topics