# Builders and engines Source: https://opvs.ai/docs/factory/builders > Hiring builders, dev agent sessions on a desk, and the engines a build can run. A **builder** is the AI employee that turns a Ready card into a pull request. It works on its own desk, with an **engine** (the coding agent that writes the code) chosen per card. ## Builders are hired Builders are products in the OPVS catalog. Hiring one puts it on your workspace with its own dedicated desk. ```bash opvs employees list # the catalog opvs employees get # one product page, with its price opvs employees hire --yes # spends a plan seat and may bill opvs employees team --agent-type builder # your builders ``` Fact Detail Kinds of hire `builder`, or `conversational` for every other AI employee. There is no separate "dev agent" hire type Needs An active subscription. Without one the hire is refused with `402` Twice A second active hire of the same builder is refused with `409`; `detail.existing_hire_id` names the one you have Cancel `opvs employees cancel ` stops it at the end of the period; it keeps working until then. There is no pause Equip `opvs employees equip --package <@vendor/name>` adds a marketplace package to a live hire **Hire or build your own?** If the catalog already has the role, hiring is one command. Build your own agent only for work that is yours alone, for example as a marketplace package under your own vendor name. ## Dev agents on a desk Besides the builder's own claimed work, a person can run agent sessions on a desk directly: open its terminal in the dashboard, or schedule prompts into named sessions with `opvs env routines` (see [Machines, desks and the sidecar](/developers/factory/machines-and-desks)). These are the sessions the [agent-ops kit](/developers/factory/agent-ops-kit) runs its planner, worker and orchestrator routines in. ## Engines The engine (also called the **harness**) is the coding agent a build runs. Each card can name one; otherwise it inherits the board's, then the project's, then the desk's. ```bash opvs engines list ``` Engine Name on cards Claude Code `claude_code` Codex `codex` OpenHands `openhands` OpenCode `opencode` Cursor `cursor` Grok `grok` Pi `pi` Antigravity `agy` Switch an engine off for the whole workspace, for example during a provider outage, and on again: ```bash opvs engines off codex --reason "provider outage" --until +2h opvs engines on codex ``` `--reason` is required and shown to the next person who looks. Without `--until`, the engine stays off until you switch it on. ## Engine and runtime are different A desk's **runtime** is the agent program a person watches in its terminal (`claude_code`, `codex` or `custom`). The card's **engine** is what an autonomous build runs. A desk on one runtime can build cards pinned to another engine, as long as that engine is installed on the desk: the desk's health lists them under `worker.harnesses`. A card pinned to an engine the desk cannot run fails within seconds of the claim. ## Talk to a running build ```bash opvs builds read # current command, recent steps and messages opvs builds tell "Use the v2 client" # send it a message, or save one on the card ``` Related: [Set up your factory](/developers/factory/set-up) step 5, [Coder profiles](/developers/guides/coder-profiles).