The language of agents.

Moving an inactive AI agent into long-term storage while preserving its state for future restoration.

The pluggable execution engine that governs how an Agenhood agent reasons and calls tools.

A collection of AI agents provisioned, run, and managed together under shared infrastructure.

The states an AI agent moves through, from creation to deletion, and the actions that move it between them.

The core observe, plan, act, and repeat cycle an AI agent runs through to complete a task.

The mechanisms by which an AI agent retains and reuses information across tasks and sessions.

The coordination layer that manages how AI agents are provisioned, scheduled, and monitored.

Bringing an AI agent out of an error state and back into normal operation.

The live, running instance of an agent: a Docker container provisioned and driven by the control plane.

The PID-1 process inside every Agenhood agent container that runs the driver and streams events out.

A reusable, packaged set of instructions that extends what an AI agent knows how to do.

A single unit of work submitted to one Agenhood agent, with a free text or schema validated output.

A preconfigured bundle of driver, model, tools, skills, and prompt used to provision an agent in one step.

The persistent /workspace volume attached to an agent, surviving restarts, pauses, and archival.

AI systems designed around planning, tool use, and iterative action rather than single-shot output.

A multi-step task in which an agent plans, acts, evaluates results, and adapts its approach.

A software system that uses a model to reason, act, and complete tasks with limited human input.

The ability to inspect and understand what an autonomous AI agent did, and why, during or after a task.

A secret token that authenticates requests to an API without a username and password.

The automatic transition of a paused AI agent back to a running state when it receives new work.

An AI agent that makes decisions and completes multi-step tasks with minimal human oversight.

A tool that gives an AI agent the ability to execute shell commands during a task.

A special address cloud instances use to query their own configuration and credentials.

An isolated environment where agent generated code runs without affecting the host system.

A keyboard-driven quick-navigation and action menu in the Agenhood console, opened with Cmd+K.

An isolated runtime, built from OS-level virtualization, that restricts what a process can see and do.

The maximum amount of text a language model can process at once, measured in tokens.

The stateless FastAPI service that owns Agenhood's public API and provisions agent containers.

A cron-based cadence that triggers a prompt or a task workflow at defined times.

A tool for defining and running a multi-container application from one declarative configuration file.

The practice of inspecting and restricting outbound network traffic leaving a system.

A network intermediary that all outbound traffic from a system must pass through.

Storing secrets such as API keys and tokens in encrypted form so plaintext is never exposed at rest.

Continuing a pretrained model's training on a smaller dataset to adjust its behavior.

An external Git repository, connected over SSH, that receives a copy of an agent's workspace after each task.

Confident but false or fabricated output produced by a language model.

Programmatically controlling a browser without a visible interface to render and read pages.

A design pattern where a person reviews or approves an AI agent's actions at defined points.

A property where performing an operation multiple times has the same effect as performing it once.

An automated policy that pauses an inactive AI agent to free up compute resources until it is needed again.

A terminal in the Agenhood console, linked over WebSocket, giving a shell inside an agent's container.

A standalone service that lets external events like GitHub issues or Slack messages trigger Agenhood agents.

A neural network trained on massive text data to predict and generate human language.

A mechanism for splitting root privileges into narrow, independently grantable units.

The console view that renders an agent's task event stream in real time as a task runs.

An AI agent instance that persists across tasks and sessions instead of being recreated each time.

A curated list of language models an application or platform makes available to choose from.

The open protocol for connecting AI applications to external tools and data sources.

A company that develops and offers access to large language models, usually via an API.

An architecture where multiple AI agents with distinct roles coordinate to complete a shared task.

An architecture where a single application instance serves multiple isolated customers or organizations.

Reverting an AI agent to a prior state without discarding the state that existed before the rollback.

An open standard that lets one application access resources on another without sharing passwords.

A task level specification defining the required shape of an AI agent's final answer.

A security extension to OAuth that protects authorization code exchanges from interception.

The practice of designing prompts to reliably get the desired output from a language model.

A security risk where untrusted input manipulates a language model into ignoring its instructions.

Restricting how many requests a client can make to an API within a given period of time.

A container configuration where the base filesystem cannot be modified while the container runs.

Constraints on how much CPU, memory, or other resources a process or container may consume.

An architectural style for web APIs that models functionality as resources accessed via HTTP methods.

A saved snapshot of an AI agent's state at a specific point in time that it can be reverted to.

A technique combining a language model with external document retrieval to ground its answers.

A server that forwards incoming client requests to the appropriate backend service.

An authorization model that grants permissions based on a user's assigned role rather than individually.

An AI agent whose actions run inside an isolated environment separate from the host system.

Software and systems an organization runs on its own servers instead of a third-party managed service.

A web search service an organization deploys on its own infrastructure instead of a third-party provider.

A web standard for streaming a one-way sequence of updates from a server to a client over one connection.

An architecture where an entire application, including all its services, runs on one virtual machine.

A model response constrained to a defined, machine readable format such as JSON.

The instructions given to an AI model before a conversation to set its role and behavior.

The ordered sequence of structured events an Agenhood agent emits while a task is running.

An ordered pipeline that chains multiple agent tasks together, possibly across different agents.

A parameter that controls how random or predictable a language model's output is.

An isolated organization, account, or workspace within a multi-tenant system.

The basic unit of text, such as a word piece, that a language model reads and generates.

A limit on how many tokens a request, task, or agent is allowed to consume.

How AI models call external functions to act on the world instead of only generating text.

Agenhood's built-in tool-use loop driver, configured directly instead of an external harness.

A dependency on a single vendor's platform that makes switching to an alternative costly or impractical.

A tool that retrieves the full content of a specific URL for an AI agent to read.

A tool that lets an AI agent query the web and retrieve results to inform its answers.

A callback mechanism where one system notifies another of an event via an HTTP request to a preset URL.

A protocol providing a persistent, two-way communication channel between a client and a server.

An automatic, restorable capture of an agent's workspace, taken after every completed task.

A network security model that verifies every connection instead of trusting internal networks by default.