Quadratique Docs
Create and manage Agents
An Agent is a reusable configuration for delegating issues. It binds a name and instructions to one detected Runtime and model, then limits where and by whom it can be invoked.
Prerequisites
Before creating an Agent:
- connect a development machine;
- confirm that the machine is online;
- confirm that it reports a supported Runtime and model; and
- identify at least one board the Agent may work on.
Create an Agent
- Open workspace Settings and then Agents.
- Select New Agent.
- Enter a nonblank name, such as Backend fixer.
- Select one available Runtime and model.
- Write the reusable instructions for the Agent.
- Select one or more allowed boards.
- Keep the Agent private or explicitly share it with the workspace.
- Create the Agent.
The Agent receives a stable identifier and starts at revision 1. Agent names are for people; routing stores the stable Agent identifier.
Write reusable instructions
Put enduring behavior in the Agent instructions: expected engineering practices, repository conventions, validation requirements, and boundaries. Issue-specific work remains in the issue title, description, and review feedback.
Do not put Runtime credentials or environment secrets in Agent instructions. Those remain on the connected machine.
Allowed boards
Every Agent must allow at least one board. Quadratique rejects routing from a board outside the allowlist, even when the member may otherwise invoke the Agent.
Share an Agent
Agents are private to their owner by default. To let workspace members delegate issues to an Agent:
- share the Agent with the workspace; and
- share the machine that hosts its Runtime.
Sharing only one of the two is not enough. The board allowlist still applies.
Edit an Agent
You can change its name, instructions, Runtime, model, sharing policy, or allowed boards. Each accepted edit increments the Agent revision.
Future runs use the new revision. A queued or active run keeps the immutable snapshot created for its attempt.
Suspended Agents
Quadratique suspends an Agent when its bound machine or Runtime is removed or becomes unavailable for use. A suspended Agent cannot receive new routing until it is rebound to an available Runtime and model.
Retire an Agent
Retirement is final. A retired Agent cannot receive new routing, and queued routing is cleared. Already-active runs may finish with their existing snapshots.