Signal is available in closed Enterprise beta. Contact your Arize account team for access.
Signal is a built-in worker that scans an LLM project’s traces on a schedule, groups recurring failure patterns into issues, and writes an investigation for each one (overview, trace evidence, suggested fix).
Requires an LLM project with traces in Arize. See Set up tracing if needed.
No repo or extra skills are required to detect issues. To let Signal open fix PRs, use Attach Repo from the Signal tab menu (⋯ on the tab header). Unattach Repo in the same menu to revert to trace-only investigations.
Enable Signal
Open the LLM project
Signal only reads traces from the selected project.
Open the Signal tab
Select Signal in the project navigation.
Click Enable Signal
Arize creates a default automation using the Signal preset (recurring schedule, for example every 10 minutes). The tab stays linked to this automation until you disable Signal or delete it.
Read issues
The Signal tab shows issue volume over time, a breakdown by entity when tagged, and a ranked issue list. Select an issue to read its investigation:
- Overview — summary of the failure pattern
- Evidence — trace IDs and filters
- Proposed fix — suggested prompt, code, config, or eval change
Use View matching traces to open evidence in Explore your traces.
Attach a repo for fix PRs
When Signal has a repo attached, it can clone the repository and open pull requests for proposed fixes—not just write investigations.
Open the Signal tab menu
On the Signal tab, open the ⋯ menu in the tab header.
Click Attach Repo
Choose an existing GitHub skill or Create new GitHub skill (name + personal access token). See Skills and permissions for token scopes. Enter the repository
Provide the repo in owner/repo form. Click Attach Repo.
Arize creates or updates the Signal agent preset for this project automatically—you do not need to visit Agent Presets first. To go back to trace-only mode, choose Unattach Repo from the same menu.