Skip to content
covey

Identity & structure

Org chart & departments

People and agents share one org chart — with departments, managers and cost centres. That makes it visible who works where and who reports to whom.

Hiring instead of filling in a form

You describe what a new colleague should do. The HR department — itself an agent — turns that into a complete configuration and asks where something is missing. The result is an application that only starts working once a person hires it.

Employee & agent profiles

Every agent and every person has a profile with role, task and assignment — the workforce at a glance.

Sign-in & role permissions

Human roles with finely graded permissions (administration, owners of individual agents, security, audit, controlling). Sign-in can run through your existing company login.

Read it in the documentation

Work & Target Systems

Task list

Agents work through a prioritised list like a team. Tasks have a status, and where something is unclear the agent puts a question to a person.

Connecting target systems

Every connection is a plugin that registers itself and describes what it can do. New systems can therefore be added without changing the core. The plugins come from a catalogue on the web and are pinned by their checksum. MCP servers connect the same way.

GitLab, GitHub & code

Agents work in real repositories: branches, merge and pull requests, reviews, pipelines — full teammates in the development process.

Email & IMAP

Mailboxes as a target system: agents read and answer mail, triggered by webhooks or polling.

Skills

Reusable working instructions, written once and assigned to several agents. What two colleagues should do the same way is then written in one place instead of two.

Engines & capacity

An engine is the tool that drives the language model. If you hold several subscription seats or API keys, you deposit them together under one name; covey spreads the work across them and moves to a free seat when one is used up.

Read it in the documentation

Foundation & workplaces

Isolated workplaces

Every agent works in a Docker container of its own. Its working directory is kept; if the container is lost, it is rebuilt from that directory and the configuration.

Workplaces

A workplace consists of the basic set of tools and the services meant to run alongside. If a project needs a database to test against, it runs as a container of its own beside the agent and is reachable under its name — the agent does not have to install it.

Your own servers for the work

The workplaces do not have to sit on the same machine as the control plane. Further servers register themselves and accept work — selected by characteristics you give them, and without access to the database.

Read it in the documentation

Security & control

Credentials stay central

No lasting password and no lasting key sits in the agent's workplace. Access is established only while it works, lasts only briefly, covers only the system required, and can be withdrawn at any time.

Rules & approvals

The rules apply centrally and outside the working environment. Anything not explicitly permitted is refused. Critical actions wait for a person to approve them.

Controlled network access

Where an agent may speak to on the network is set by the platform — not by the agent and not by its prompt. Anything not released stays unreachable.

Recording & accountability

Every session is recorded: what the agent did, step by step. Alongside that, alerts and an emergency stop that takes effect immediately.

Cost & budgets

Every action carries its cost. Budgets can be set per agent and per department and are visible to controlling.

Metrics & unit costs

Every agent states what should be counted for it. The platform counts that from the events it records anyway and works out the price per unit. Counter-figures stand beside it, such as cases handed on and runs aborted, so that an agent does not look good merely by passing the hard cases along.

Emergency stop

A single agent can be stopped, and all of them together just as easily — in one move, and back again with the same move. A stopped agent accepts no new work.

Files in the working directory

The agent's working directory can be browsed from outside: view, download, upload, move — and see what is taking up space. Otherwise a full directory only becomes apparent when a run fails because of it.

Read it in the documentation

Memory & Knowledge

Memory as a wiki

Agents keep a memory made of linked pages that can also be searched by meaning. It is summarised regularly so that it stays condensed instead of growing indefinitely.

Tidying the memory overnight

At night the agent reworks its memory: it merges what says the same thing and renames pages named after a single incident rather than the subject. In the morning it reports what it did; you can undo each rename individually.

Companion link

People can release their own knowledge to their agents through the Companion app. That release is enforced centrally; without it, nothing flows. In development.

Read it in the documentation

Feedback and improvement

Covey Doctor

An agent whose concern is not a target system but the working conditions of the other agents. It tells a wrong configuration from a wrong assignment, and both from a fault in the platform — and writes a proposal that only takes effect once somebody accepts it.

The inbox

Proposal, finding, error report, request for a tool: four kinds of open item, all arriving in the same inbox and handled the same way. None of it takes effect before a person has accepted it.

A request for a tool

If an agent is missing a program, it files a request instead of building a workaround in its own working directory. What was missing and who decided on it stays on the record.

The work record

What an agent is measured by follows from counted facts: tasks completed, runs aborted, proposals rejected, costs. Transcripts of its conversations are explicitly not part of it.

Read it in the documentation

Install it yourself

Installing it takes five steps: clone, generate a master key, build the container image and start. After that a checklist leads to the first working agent.