Self-hosted AI agent stacks
LLM agents that call tools, keep memory and run multi-step workflows, deployed on infrastructure you control.
Agent stacks
- AI Agent Framework (agent) – Framework for building autonomous AI agents with tools
About Agent stacks
An agent stack lets a language model decide which tools to call (APIs, databases, search, code) and loop until a task is done. Self-hosting matters here because agents often touch internal systems and credentials that should not leave your network.
Before you give an agent real permissions:
- Use a model that handles tool calling reliably, and validate every tool argument
- Run tools with the least privilege they need, ideally sandboxed
- Treat retrieved content and tool output as untrusted input (prompt injection)
- Log every step so you can see why the agent did what it did
Frequently asked questions
- What is a self-hosted AI agent?
- An AI agent whose language model, tools and memory run on your own servers. The model decides which tools to call and in what order to complete a task.
- Can local models do tool calling?
- Yes. Several open-weight model families support tool calling, and servers such as Ollama and vLLM expose it through an OpenAI-compatible API. Reliability varies by model, so test with your own tools.
- How do I keep an AI agent safe?
- Give tools minimal permissions, require confirmation for destructive actions, sandbox code execution and treat all external content as untrusted, because it can contain instructions aimed at the model.
Guides
- Self-Hosted AI Agents: Local LLM Tool Calling That Works – How tool calling works on Ollama and vLLM, which open-weight models handle it, and how to wrap it in an agent loop that is safe to point at real systems.