Use Hands with Intelligence

Move an agent to Veyra Hands, steer and stop its runs, and read what it did, with budgets and approvals unchanged.

Veyra Docs TeamUpdated 3 min readVersion 1 · History

In Veyra Intelligence every agent has a harness and a runner. With Veyra Hands as the runner, each of the agent's runs goes through Hands, while Intelligence keeps deciding what the agent works on, what it may do and what it may spend. This guide covers the day-to-day: moving agents, steering runs and reading what happened.

Move an existing agent to Hands

An agent hired before Hands, or on an environment that did not support it then, keeps the earlier runner until you move it.

  1. Open the agent and go to its Runtime tab.
  2. When Hands can run this agent, the tab shows Upgrade to Veyra Hands. Select Upgrade runner.
  3. Intelligence checks the agent's environment first. On a machine it has not checked before, this can take a few minutes.

The agent keeps its harness, model, account and compatible settings. Its next session starts fresh; past runs and their transcripts stay where they were. If a setting does not carry over, the tab says which one before you upgrade.

Choose the runner for a new agent

When you hire an agent, Runner starts on Automatic: Intelligence uses Hands whenever the agent's environment supports it for the chosen harness. Choose Veyra Hands to require it, so the agent is never hired onto the earlier runner by accident. Cursor Cloud agents stay on the earlier runner unless you choose Hands, because Hands support for Cursor Cloud is experimental.

Follow a run

Open the task the agent is working on. While Hands runs the turn, the thread shows:

  • the agent's messages as they stream;
  • each tool call with its input and result;
  • file changes in the agent's workspace;
  • questions and approval requests that wait for you.

The task's run console shows the same run with its full event log. When the run ends, a readable summary sits on top and the transcript stays beneath it.

Steer, stop and answer

  • Steer: post a comment on the task while the agent is working. Hands delivers it into the running turn and the agent reads it at its next step. If the run restarts before the model has read it, Hands delivers it again.
  • Stop: select Stop run in the task's thread. Hands ends the turn and the harness process, and the run is recorded as cancelled.
  • Answer: when the agent asks a question or a governed action needs approval, it appears in the task and in your Inbox. The agent continues once you answer.

When a connection drops

Hands saves checkpoints while it works. If the connection between Hands and Intelligence drops, the run reconnects and continues from where it was, and a tool call that already ran is not run a second time. When the harness cannot resume its own conversation, Hands starts a fresh one and hands it a short summary of the task's history: the original request, recent messages and the answers you gave.

When you change the agent's tools between runs, for example by assigning it a new connector, harnesses that can refresh their tools keep their conversation; the others start the next session fresh.

Tools, permissions and budgets

Hands changes how a run is carried out, not what the agent is allowed to do:

  • Every tool call goes back to Intelligence and is checked against the agent's permissions before it runs.
  • Governed actions still wait for approval.
  • Runs on Hands count against the agent's and the organization's budgets, and work pauses at the limit as before.
  • Every change the agent makes lands in the activity log, attributed to the agent.

Next steps

AgentsHandsRuns