An open-source agent fleet, defined by its rules
An agent fleet is a group of AI agents managed as one population, with a shared registry, a shared policy and one record of what ran. Kortix is the open-source CompanyOS: your agents, skills, memory, and connectors in one repo you own.
What a fleet's rules look like in a file
# yaml-language-server: $schema=https://kortix.com/schema/kortix.schema.jsondefault_agent: assistantagents: assistant: description: Answers questions about the project. prompt: You answer questions about this project. Read the files before you answer. tools: all # omitted: all, with a warning skills: all # omitted: no skills workspace: { project_repository: true } # omitted: no repository in /workspaceOne manifest names the default agent, its prompt, its tools and its skills, and every other key stays optional.
A fleet, or just agents running
Kortix keeps the registry, the policy and the record as files in a repo the company owns, and each column shows where the other two put them.
Where the definitions live
Who approves a write
Audit record
Self-hosting
Open source
Each rival column states what that product documents on its own page, checked November 2026.
What an agent fleet is
An agent fleet is the managed population of AI agents a company runs, taken as one unit: every agent in it, the rules they share, and the record of what they did. The phrase names a management job, so it survives a change of model or vendor.
A single agent is easier to describe. An agent is an assistant with a job description. It has instructions that say what it does, a model it thinks with, and permissions that decide what it may touch. That definition says nothing about who else is running, and fleets begin there.
A fleet adds the questions that appear once several agents work at once. Who owns each one. What may each touch. Who approves a write, and where the approval lands. Five agents with one registry, one policy and one audit trail are a fleet. Ten agents with none are ten scripts that happen to run on the same laptop.
Four things that make a population a fleet
A registry
Every agent in a fleet has an entry: a name, an owner, and the job it is meant to do. The registry answers who asked for this agent and what it is for without opening a chat window. A registry kept in one place also gives every agent the same shape. The sixth agent becomes a file someone can review.
Shared policy
A policy says what an agent may touch: which connectors, which files, which commands. Without one, permission belongs to whoever configured the agent last. Kortix keeps the rule beside the agent. A connector rule set to Ask holds the action until a person approves it.
Approvals
An approval is the moment a person says yes to a specific action, and the fleet has to remember it. An approval covers one exact action. A changed detail needs a new approval. Without that record, a rule everyone assumes somebody checked becomes a rule nobody checked.
An audit record
The record answers what ran, for whom, and at what cost. It is also the part a finance team asks for first. Kortix records every action in an audit log, and each entry has a title and a route. A fleet that cannot produce that record cannot answer a customer's question about last Tuesday's change.
When a population is not a fleet
Three agents that never touch the same systems are one script each. A script with a schedule and a prompt is a reasonable thing to own. Calling it a fleet adds a word and no control.
Twenty agents on one platform with no registry form a queue. Work arrives and some agent picks it up. Nobody can say which one ran it, or who to ask when the output is wrong. Volume is no test in either direction.
Four questions settle the case for a real fleet of agents. An afternoon is enough to answer them:
- Who owns this agent, by name?
- What may it touch, written down?
- Who approved its last write, and where is that recorded?
- What did it do last week, and what did it cost?
When those answers live in four different heads, the company has agents and does not yet have a fleet.
The other things called a fleet
Two other products use this word for something else, and the difference decides which tool a device team needs.
Elastic Fleet is Elastic's management layer for Elastic Agents inside Kibana, and those agents collect logs and metrics from hosts. Elastic Fleet does not run AI task agents; it moves observability data.
Fleet (fleetdm.com) is open-source mobile device management for macOS, Windows, Linux, iOS and Android. "Fleet" here means devices, not AI agents.
Neither product runs the work agents that make up an agent fleet. A search for device management belongs on their pages.
What a company owns when it runs a fleet
A fleet is a management surface before it is a set of agents. The company has to own that surface. Where the registry lives decides who can read it and who can change it. It also decides what happens when a vendor changes its pricing.
Kortix keeps the surface in a repo the company owns. A project is one git repo. Its agents, skills, memory and connector settings are files in that repo. When an agent changes the project itself, for example a skill or its memory, it opens a change request. A person reviews it on the Review page and merges it.
Kortix sets that arrangement against FleetQ and LangSmith Fleet: the registry and the policy are files you can read and diff, the code is open source, and the same stack runs on your own machine.
Agent fleet questions, answered plainly
- What is an AI agent fleet?
- An AI agent fleet is a managed group of AI agents that share one registry, one policy, one approval record and one audit log. Five governed agents are a fleet; ten ungoverned ones are separate scripts.
- What does "Fleet Agent" mean?
- A Fleet Agent usually means the software a device-management product installs on a machine: Elastic Fleet manages Elastic Agents that ship logs and metrics from hosts, and FleetDM enrolls laptops and phones. Neither kind runs AI work agents.
- What is the purpose of a fleet?
- The purpose of an agent fleet is to run many agents under one set of rules: who owns each agent, what it may touch, who approves a write and what record is left. A fleet gives a company one place to answer those questions.
- Is ChatGPT an agent or LLM?
- ChatGPT is a product built on an LLM. The model answers text, and the product around it adds tools, memory and a chat interface, which is what lets it act as an agent. In a fleet, the model is one part of an agent's setup.
Write the rules down before the second agent arrives.
Start with one agent, then govern the population it grows into.
Every action is recorded · Self-host, VPC, or on-prem · Any model, your keys