Skip to main content
A managed agent does investigation and repair work for you in an Arize-managed sandbox, using the traces, monitors, and repositories you already connect to Arize AX. You can hand it a one-off job such as investigating failing traces or fixing a bug in your repo, or put it on a schedule or metric threshold so it triages a monitor alert without you starting it.

Start from a ready-made agent

Every managed agent starts from Signal & Agents → New Agent. Pick a ready-made agent and adjust it, or start from scratch. The gallery groups agents by purpose.

Start from scratch

Under Start from scratch, select Blank Agent to open Agent Studio with nothing filled in.

Guided setup

A ready-made agent walks you through up to three steps before Agent Studio. Steps the agent does not need are skipped, such as Connect for an agent that reads only Arize traces.
1

Connect

Connect outside systems such as GitHub (clone a repo, open PRs) or Dynatrace (metrics and logs). A GitHub skill adds a repository field. Arize trace access is provisioned automatically and never appears here.
2

Configure

Select the tracing project the agent reads, and choose a one-off session or a recurring automation. An automation needs a name and a trigger.
3

Review & launch

Edit the task prompt, confirm the setup, and start it.
The first run may ask you to acknowledge that the harness processes project traces, your instructions, and connected repo or tool data.

Run a session or set up an automation

  • Run session starts a one-off run. You can follow the run and send messages to it. A session can be configured ad hoc without an agent config.
  • Setup as automation runs the same configuration on a Cron schedule or a Metric threshold. Automations require an agent config.
Some agents support one mode: Fix a bug in your repo is session-only, and Triage a monitor alert and Recurring health check are automation-only. Signal uses the same automation model, enabled per tracing project.

Customize a run

Agent Studio lets you set these options for a run. AI providers and skills are covered in Connect your harness.

Save an agent config

An agent config bundles a harness, sandbox, and optional AI provider, skills, and repository for reuse across runs and automations. Agent configs are managed under Settings → Space Settings → Managed Agent Configs. See Managed Agent Configs & Skills.
  • Attaching a repo in Configure Signal creates a Signal-managed agent config, hidden from the general list.
  • The project is picked at launch, so one config serves many projects.

Track runs

Open Signal & Agents and select Agents. Choose Grid view to see runs grouped by agent config and colored by status, or List view to see sessions and automations in a searchable table. From a run you can read the live transcript, send follow-ups, open a PR the agent created, view the agent’s own traces, reprovision a finished session to continue it, or archive it. Multiple sessions run in parallel. For an automation, View automation details shows its trigger, prompt, agent config, recent runs, and any PRs it produced, and lets you pause, resume, or delete it.

Edit a custom automation’s prompt

Edit prompt is available on custom automations you have permission to manage. Signal and other issue-producing agents have platform-managed prompts.
  1. Open Signal & Agents, select Agents, and switch to List view.
  2. Select View automation details on the automation’s row.
  3. Open the More actions menu at the top right of the automation panel, not the menu inside a run, and choose Edit prompt.
  4. Update the Prompt field and select Save. The prompt cannot be empty.
Future runs use the new prompt. Runs in progress keep the old one.

Outputs

Code changes arrive as a pull request to review and merge. A managed agent never deploys to production.