AI agent
Software that uses one or more language models to autonomously plan and execute multi-step tasks via tool calls.
50 terms you'll meet again and again working with self-hosted AI agents. Concise definitions, with cross-references to related concepts and to our deeper guides.
Software that uses one or more language models to autonomously plan and execute multi-step tasks via tool calls.
Deployment with no network connection to untrusted networks — typically no internet access at all.
UI pattern requiring human confirmation before an agent's tool call executes.
Pricing pattern where a user supplies their own LLM API keys to a self-hosted or managed agent.
Reverse proxy server with automatic HTTPS via Let's Encrypt. The easiest path to TLS on a self-hosted setup.
Edge serverless runtime running V8 isolates across 330+ Cloudflare data centres.
Critical 1-click remote code execution vulnerability in OpenClaw versions before 2026.2.10. CVSS 9.6.
Regulatory or contractual requirement that data not leave a defined geographic or organisational boundary.
Tool for defining and running multi-container Docker applications via a single YAML file.
Sandbox using a Docker container with filesystem mount and network policy controls.
AI workloads running close to where data is generated rather than in centralised cloud datacenters.
Numerical vector representations of text, images, or audio that capture semantic similarity.
EU regulation classifying AI systems by risk level and imposing obligations on high-risk deployments.
EU regulation governing personal data processing. Applies to most consumer-facing AI deployments touching EU users.
File format for storing quantised LLM weights, used by llama.cpp and Ollama.
User-space kernel that runs as a sandbox layer between containers and the host kernel.
Open-source self-hosted AI agent from Nous Research, released February 2026. Sandboxed by default, multi-LLM.
German VPS and dedicated server provider. The cheapest credible EU option for self-hosting in 2026.
Enterprise-grade self-hosted AI agent with gVisor sandboxing, RBAC and audit logging. Source-available.
Meta's family of open-weight large language models. Llama 3.3 70B is the leading open frontier model in 2026.
C++ inference engine for LLMs. The runtime under most local-LLM setups, including Ollama.
Language model running on local hardware rather than via cloud API. Llama, Qwen, Mistral local variants.
Passwordless authentication where the user clicks a one-time link sent to their email.
Open standard for connecting language models to external tools and data sources via a JSON-RPC interface.
Process that exposes a set of tools to MCP-compatible AI agents over a JSON-RPC interface.
European LLM developer. Mistral 7B is the canonical small open-weight model; Mistral Small 22B targets the mid-tier.
Cloudflare Workers-based self-hosted AI agent reference implementation.
4,000-line Python self-hosted AI agent from HKU. Designed to be auditable in an afternoon.
macOS-only fork of OpenClaw using Apple's container framework for sandboxing. Claude-only.
Local LLM runtime that exposes an OpenAI-compatible API over local model weights.
Open-source autonomous AI agent created by Peter Steinberger in November 2025. Most-installed self-hosted agent of 2026.
LLM API gateway providing a unified interface across Anthropic, OpenAI, Google, Mistral and many open models.
Microsoft's small-language-model family. Phi-3 mini 3.8B runs on a Raspberry Pi 5.
Self-hosted AI agents and language models running on portable, low-power hardware you own.
Attack technique where adversarial input embedded in a document or webpage hijacks an agent's behaviour.
Reducing the numerical precision of model weights to shrink memory footprint and speed up inference, with quality tradeoff.
Alibaba's open-weight LLM family. Qwen 2.5 Coder 7B is widely used for self-hosted code-focused agents.
Pattern of augmenting LLM responses with retrieved external data, typically from a vector database.
Vulnerability class where an attacker can execute arbitrary code on a target system over a network.
Isolation layer that constrains what an agent's tool execution can access on the host.
Running an agent's tool calls inside an isolated environment that limits filesystem, network and process access.
AI software you install on hardware you control, rather than consuming as a hosted product.
Mesh VPN built on WireGuard. Common pattern for keeping self-hosted dashboards off the public internet.
Structured invocation of an external function by a language model, typically as JSON conforming to a declared schema.
Database optimised for similarity search over high-dimensional embeddings.
Virtual private server — a virtualised slice of a physical server, typically rented by the month.
Container that watches running containers and pulls/redeploys when new images ship.
Server-side check that an incoming WebSocket connection comes from an allowed origin domain.
Security architecture that assumes no implicit trust and verifies every access request.
Privacy-first self-hosted AI agent. Local LLMs only, no cloud, AGPL-3.0.