Sentry

Sentry can start an investigator when an issue alert fires or a metric alert enters critical status. Connecting Sentry, configuring its alert rules, and choosing an investigator are separate steps.

The Sentry integration is not generally released yet. If it is enabled for your organization, use the setup below. Contact Support for access.

Connect and configure alerts

Connect Sentry from the Ellipsis integrations settings and complete the installation in your Sentry organization.
In Sentry, configure each project's issue or metric alert rule to notify the Ellipsis integration. Installing the integration alone does not send alerts.
Merge sentry.yaml at the root of your organization's .ellipsis repository. Ellipsis syncs it from the default branch.
Trigger a test alert from Sentry, then check that an Ellipsis session starts and delivers the requested report.

The installation and token credentials stay in the integration settings. sentry.yaml controls routing and investigator behavior; it does not install Sentry SDKs, create Sentry alert rules, or store credentials. Slack reports also require a connected Slack workspace and channel access.

Route alerts

This example investigates every project in one Sentry organization and sends one short Slack report per alert. Each report identifies its project.

YAML

ellipsis:
  version: v1
  kind: sentry
  name: Production error reports

sentry:
  organizations: [my-sentry-org]
  projects: ['*']
  on: [issue_alert, metric_alert]

session:
  harness:
    type: codex
    model: gpt-5.6-terra
  instructions: |
    Investigate the alert and send one report to #eng-ops, at most
    three short lines and 70 words. Include the project, Sentry link,
    impact, and next step. Link the causing PR only when supported by
    evidence. Tag its author only after verifying their Slack identity.
    If attribution is uncertain, say so. Do not open a fix PR.
  environment:
    repositories:
      - name: api-repo
    mcp_servers:
      - name: slack
  permissions:
    github:
      permissions: read_only
  budget:
    session: 2

The routing file supports these fields:

Field	Behavior
`ellipsis.enabled`	Defaults to `true`. Set `false` to stop Sentry investigators.
`ellipsis.name`	The investigator's name.
`sentry.organizations`	Sentry organization slugs; defaults to `['*']`.
`sentry.projects`	Required, nonempty project slugs, or `['*']` for every project.
`sentry.on`	`issue_alert`, `metric_alert`, or both (the default).
`session.harness`, `session.instructions`, `session.skills`	The investigator's model, instructions, and skills.
`session.environment`, `session.permissions`	The investigator's environment and credential permissions. An environment can be inline or reference a saved environment.
`session.budget.session`	Optional positive session budget in dollars.
`session.budget.day`, `week`, `month`	Spend limits across this handler's sessions over trailing 1-, 7-, and 28-day windows, in dollars.

The file defines exactly one investigator. An alert must match its organization, project, and alert type. List several organizations or projects to use the same investigator for all of them. Do not mix '*' with names in the same list. An alert with an unknown project can only match a project wildcard. Unmatched alerts do not start a session.

A project called backend in one Sentry organization does not match a filter scoped to another organization. Session settings follow the same structure as automation YAML.

Instruction files, skills, and harness settings without an explicit repository resolve from the .ellipsis repository. Qualify a reference with repository to read it from an application repository instead.

Delivery and existing automations

When a live sentry.yaml exists, it controls all Sentry dispatch, including when disabled or when no route matches. Sentry-triggered automations (trigger.sentry) do not also run. Organizations without the file keep their existing automation behavior. Deleting the file restores that automation dispatch; use ellipsis.enabled: false when you want to stop investigations.

Invalid edits preserve the last successfully synced configuration and record a sync error. The file is only active at sentry.yaml in the .ellipsis repository; copies in application repositories are inert.

Issue alerts have a six-hour cooldown after an investigation starts. Metric alerts only start investigations on entry into critical, not on warning or resolved. Repeated deliveries are deduplicated. A live test alert is required to verify delivery across Sentry, Ellipsis, and Slack; a successful installation callback alone does not verify that path.

Connect and configure alerts

Route alerts

Delivery and existing automations

On this page