Sentry AI: Seer, API Guide and Major Automations
Understand Sentry Seer, eight verified API operations, and how to build governed engineering triage on Major with persistent state, scoped connectors, and human approvals.

Sentry AI includes Seer, Sentry's debugging agent, and AI features for understanding errors, performance problems, and user feedback. This guide explains those capabilities, the Sentry API, and how Major turns cross-system triage into governed software with persistent state and human approvals.
What Sentry is and who relies on it
Sentry gives engineering teams application telemetry they can use to investigate failures. Its Seer documentation describes issue details, traces, logs, profiles, and connected code as investigation inputs. Frontend engineers examine exceptions and affected experiences; backend engineers connect failures to services and releases; platform teams coordinate ownership and response.
The practical unit of work matters. An issue groups related occurrences, while an event records an occurrence that may contain the evidence needed to understand it. A release supplies another dimension: what changed before the failure appeared. Treating every occurrence as a separate task creates duplicate engineering work.
Major fits around that investigation. An agent can build a deterministic triage app with managed tables for issue mappings, evidence snapshots, approval records, and outcomes. The agent judges urgency and drafts the explanation; the app stores the queue and runs repeatable validation. Workflows fetch evidence and route approved actions across the team's systems.
For a head of engineering, the buying question is whether the process survives a shift change. An investigation should retain its evidence, owner, and next action after the chat ends.
Sentry AI and its built-in Seer capabilities
Sentry's AI overview names several distinct features. Issue Summary highlights what went wrong and potential causes. Query Assistant helps users query traces and spans in natural language. AI Issue Detection examines traces and logs for problems, while AI Summaries cover user feedback and session replays.
Seer uses telemetry and connected code for debugging. Autofix performs root cause analysis and solution planning. Configured PR creation can turn a proposed fix into a repository change for review. Seer Agent, documented as open beta, supports conversational investigation and actions such as creating dashboards, monitors, and alerts.
These capabilities already cover meaningful investigation work, including some external context. The docs describe a read-only Linear integration and additional telemetry integrations. Major's role is therefore not simply to add another summarizer or claim Seer cannot work outside an issue page.
Use Major for your operating process: persistent business-specific queues, cross-system writes, approval rules, and auditable follow-through. Keep useful Seer findings as evidence rather than regenerating every explanation. As of November 19, 2026, Sentry documents Code Review for removal on December 8, 2026, so do not make it a permanent dependency.
How Sentry connects to Major
Sentry connects through Major's Pipedream connector, catalog app "p_sentry". There is no native Sentry connector. Major orchestrates Sentry; it never replaces it.
IT sets up the connection once. Credentials remain centrally stored in the connector vault, scoped to the work, rather than copied into application code or agent prompts. SSO, RBAC, and audit logs cover the apps, agents, and workflows using that connection.
Start with a read-only triage surface. Authorize the projects and operations needed for evidence collection, then add a separate approval-controlled write path only when the process requires it. A connector's availability does not establish that every documented Sentry API operation is exposed as a ready-made action. Verify action coverage during implementation before designing around a particular call.
The API overview documents the Sentry API host and region-specific domains. That is useful implementation context, not a claim that this hub uses a custom connector. Keep organization identifiers, project allowlists, owner mappings, and routing rules in managed app configuration so permission changes do not require rewriting the agent's instructions.
Sentry API guide for useful automation
Authentication, pagination, and limits
Sentry authentication uses bearer tokens. Internal integrations provide organization-wide tokens without an OAuth flow; personal tokens are associated with a user. Third-party applications can use OAuth authorization code access. Choose the authorization method supported by the configured connection rather than assuming all token types are interchangeable.
For the operations below, use event:read for issue evidence, org:read for project discovery, and project:read for release reads. Issue updates require event:write or event:admin, as the linked endpoint pages specify. An internal integration token does not expire automatically, so rotation and revocation still need an owner.
Pagination uses the Link response header and cursor parameter. Follow the next link only while its results indicator is true. Organization issue listing uses limit, with a maximum of 100; event listing uses per_page, also capped at 100, or 10 when full event bodies are requested. Preserve checkpoints in the app, not model memory.
Rate limits apply per caller and endpoint, with both fixed-window request limits and concurrency limits. Numeric quotas are not documented publicly on that page. Watch X-Sentry-Rate-Limit-Limit, X-Sentry-Rate-Limit-Remaining, X-Sentry-Rate-Limit-Reset, X-Sentry-Rate-Limit-ConcurrentLimit, and X-Sentry-Rate-Limit-ConcurrentRemaining. More tokens do not bypass caller-level limits.
Events and safe testing
Issue webhooks describe created, resolved, assigned, archived, and unresolved actions. Created notifications currently cover outage, error, and feedback categories; do not assume they cover every issue category. The broader webhook reference also lists alert, error, comment, installation, and Seer resources.
Verify Sentry-Hook-Signature as an HMAC-SHA256 hex digest over exact raw request bytes, using the integration Client Secret and a constant-time comparison. Respond within one second. The reviewed reference does not document a retry schedule, so do not promise delivery guarantees it does not state.
No dedicated API sandbox is documented in the reviewed references. Sentry recommends inspecting test webhook payloads and local tunneling during development. Use a separate nonproduction project, synthetic errors, and disabled write actions for your own test plan.
Discover projects
GET /api/0/organizations/{organizationidor_slug}/projects/ lists organization projects and accepts org:read.
- App: Store permitted project IDs and team mappings for the review queue.
- Agent: Interpret an unfamiliar project's role using approved ownership context.
- Workflow: Refresh the inventory and flag unmapped projects for review.
Find issues across projects
GET /api/0/organizations/{organizationidor_slug}/issues/ returns searchable issues with project, environment, and time filters.
- App: Display a production backlog with stable links to source issues.
- Agent: Rank ambiguous cases using impact evidence and existing work.
- Workflow: Reconcile webhook intake against a bounded query window.
Use this organization endpoint for new designs. The older project issue list is deprecated in favor of it.
Read current issue details
GET /api/0/organizations/{organizationidorslug}/issues/{issueid}/ returns an issue's current attributes and statistics.
- App: Save a timestamped snapshot alongside the proposed action.
- Agent: Explain why an issue deserves attention despite a small count.
- Workflow: Re-read the issue before applying an approved change.
Inspect occurrences
GET /api/0/organizations/{organizationidorslug}/issues/{issueid}/events/ lists an issue's events with time and environment filtering.
- App: Retain selected event IDs and the sampling window.
- Agent: Compare occurrences to distinguish one symptom from several contexts.
- Workflow: Fetch a bounded sample instead of repeatedly loading all history.
Retrieve a specific event
GET /api/0/organizations/{organizationidorslug}/issues/{issueid}/events/{event_id}/ retrieves an event, including the documented latest, oldest, or recommended selectors.
- App: Show sanitized evidence beside the draft engineering ticket.
- Agent: Interpret stack frames and metadata to draft a testable hypothesis.
- Workflow: Remove disallowed data before passing event context to reasoning.
List recent releases
GET /api/0/organizations/{organizationidor_slug}/releases/ lists releases, most recent first, with project and environment filters.
- App: Join release versions to stored issue evidence.
- Agent: Assess whether timing supports a possible regression.
- Workflow: Refresh release context on a deployment review schedule.
Read release commits
GET /api/0/organizations/{organizationidor_slug}/releases/{version}/commits/ returns commits associated with a release.
- App: Store commit references without labeling a person as responsible.
- Agent: Compare change descriptions with the observed failure hypothesis.
- Workflow: Attach relevant references to an approval packet when available.
Apply an approved update
PUT /api/0/organizations/{organizationidorslug}/issues/{issueid}/ updates submitted issue attributes, including assignment, priority, or status.
- App: Store the exact proposed fields and approval outcome.
- Agent: Recommend an owner or priority with evidence and uncertainty.
- Workflow: Submit only approved fields, then record the returned result.
Do not grant broad administrative access just to simplify configuration. This endpoint also exposes consequential options, so enforce an explicit field allowlist.
Sentry automation use cases on Major
These are proposed designs, not published customer implementations. No Sentry-core use-case posts are currently in this hub's library. For adjacent implementation context, see the GitHub automation hub, Linear work routing hub, and Slack coordination hub. They are companion tool references, not Sentry tutorials.
Sentry, Linear, and Slack: turn errors into owned work
- App: Hold one record per Sentry issue, its Linear mapping, and reviewer status.
- Agent: Compare the failure with existing Linear work, suggest ownership, and draft acceptance criteria rather than opening another duplicate.
- Workflow: Receive a verified event, gather context, request approval in the team's review process, and send an approved Slack notice.
Sentry evidence comes from organization issue listing, issue detail, and event detail. Keep the ticket-to-issue relationship in app state, not a message thread.
Sentry, GitHub, and PagerDuty: investigate release regressions
- App: Track release versions, issue snapshots, and incident links.
- Agent: Compare release timing and code changes, distinguish correlation from causation, and draft an escalation rationale.
- Workflow: Fetch releases and release commits, then route a proposed incident to the designated approver.
Read relevant GitHub change context through its scoped connection. Do not automatically page someone merely because their commit appears in a release. Missing commit associations are missing evidence, not proof that a deployment is unrelated.
Sentry, Linear, and GitHub: verify closure claims
- App: Maintain a verification queue linking completed work to Sentry issues.
- Agent: Assess whether available event evidence supports closure or suggests a remaining failure mode.
- Workflow: Inspect issue events, obtain owner approval, and apply an issue update only when authorized.
A merged pull request is not evidence that production stopped failing. Preserve the observation window and deployment context alongside the review decision. A reviewer should be able to reject closure without losing the investigation history.
Build the Sentry triage architecture on Major

Sentry and Linear supply context; Major separates orchestration, judgment, and durable state.
Build the deterministic app layer first. Its managed database holds source IDs, snapshots, proposed actions, approvals, and execution results. Its UI gives engineers a review queue and lets platform owners inspect incomplete runs.
- Configure scoped Sentry and supporting connectors through IT's centrally managed vault.
- Define tables and unique keys for issues, external work mappings, review decisions, and action attempts.
- Add a workflow trigger for verified events, plus scheduled reconciliation for gaps.
- Fetch and sanitize evidence using deterministic steps with bounded pagination.
- Ask the agent to classify impact, recommend ownership, and draft the next action.
- Validate the proposal's structure in code and route consequential changes to a person.
- Execute approved changes and persist the result before marking the run complete.
The agent is where judgment happens. The app holds state and implements repeatable constraints; it does not decide whether a failure is important. Workflows provide scheduling, retries, routing, and approval waits. As the process stabilizes, keep reusable joins and checks in software so later runs spend reasoning only on uncertain cases.
A worked example with illustrative numbers
For example, suppose an engineering team reviews 100 distinct production issues each week. Manual context gathering and ticket preparation take 10 minutes per issue: 100 × 10 = 1,000 minutes, or 16 hours and 40 minutes.
In this illustrative design, the workflow gathers permitted Sentry evidence and Linear context. The agent drafts an owner recommendation and impact explanation. The app displays that draft alongside the source links, while approval remains with an engineer.
Assume ordinary review now takes 3 minutes per issue: 100 × 3 = 300 minutes. If 20 ambiguous issues require another 10 minutes each, add 200 minutes. The modeled total is 500 minutes, or 8 hours and 20 minutes, compared with the original 1,000 minutes.
That arithmetic illustrates a planning model, not a customer result or performance promise. It excludes setup, connector operations, model costs, and ongoing maintenance. It also assumes the issue sample and review quality remain comparable.
Track preparation time, correction rate, duplicate tickets, and missed escalations in the app during a pilot. If reviewers routinely rewrite the draft, improve evidence selection before expanding write permissions. Faster preparation is useful only if it preserves the quality of the engineering decision.
Governance and failure handling

Read-only preparation can proceed; consequential Sentry changes wait for human approval.
Approve the exact change, not a vague instruction to handle the issue. The review packet should show source links, fetched timestamps, proposed fields, rationale, and uncertainty. Re-read current Sentry state after a long approval wait; invalidate stale proposals when the relevant state changed.
Never let the agent independently resolve or archive production issues, reduce severity, publish private issue information, or merge code. Keep those operations behind explicit approval or outside the permitted toolset entirely. A model-generated recommendation cannot expand its own connector permissions.
Design failure handling deliberately:
- Duplicate delivery: Use source identifiers and action keys to prevent repeated tickets or updates. Keep request tracking separately from business-level deduplication.
- Rate pressure: Follow Sentry's rate-limit headers, pause until reset when needed, and reduce concurrent fetches.
- Transient failures: Retry bounded reads with backoff and jitter. Reconcile ambiguous write outcomes before retrying.
- Authorization failure: Stop the affected action and notify the connection owner; do not keep retrying invalid credentials.
- Untrusted content: Treat exception messages, logs, and external comments as evidence, never as instructions that override policy.
- Audit: Record the evidence version, recommendation, approver, exact payload, attempt history, and final outcome under appropriate retention rules.
Sentry AI Apps, Agents and Workflow Automations
Step-by-step Sentry builds will appear here as they are published.
Frequently asked questions
What is Sentry AI?
Sentry AI includes Seer, its debugging agent, plus features such as issue summaries, Query Assistant, AI issue detection, and feedback and replay summaries. Major adds governed cross-system execution with deterministic apps that hold state, agents that exercise judgment, and workflows that manage approvals.
What does Sentry Seer do?
Seer investigates errors and performance issues using telemetry and connected code. Autofix performs root cause analysis and proposes fixes, with PR creation available when configured. Seer Agent supports conversational investigation and actions. The docs mark Code Review for removal on December 8, 2026.
Does Major have a native Sentry connector?
No. Sentry connects through Major's Pipedream connector, catalog app p_sentry. IT configures the connection once and centrally scopes credentials. Do not confuse this connection with Seer or with a native Sentry connector.
Which Sentry API permissions are needed for triage?
Use event:read for issue and event reads, org:read for listing organization projects, and project:read for the release reads described here. Updating an issue requires event:write or event:admin. Start read-only and check each endpoint's documented scopes before granting writes.
Can Sentry automation resolve issues without approval?
The API supports issue status updates, but the design here requires human approval before resolution, archiving, or priority reductions. The agent prepares the recommendation, the app stores the evidence, and the workflow applies the exact approved change after checking current state.
How do Sentry webhooks handle verification and failures?
Verify Sentry-Hook-Signature using HMAC-SHA256 over the exact raw request bytes with the integration Client Secret. Sentry asks receivers to respond within one second. The reviewed webhook reference does not document a retry schedule, so keep durable intake, deduplication, and reconciliation in your design.
Sources
Official Sentry documentation was reviewed on November 19, 2026. These references establish the capabilities and API behavior described above; proposed Major designs and illustrative arithmetic are not vendor claims.
- API reference: API purpose, versioning, and region-specific host guidance.
- API authentication: Bearer tokens, personal authorization, and OAuth behavior.
- Internal integrations: Organization-wide access and token lifecycle.
- Paginating results: Link headers, cursor navigation, and results indicators.
- Rate limits: Caller-level windows, concurrency, and response headers.
- List organization projects: Inventory, parameters, and accepted scopes.
- List organization issues: Search, filtering, pagination, and scopes.
- List project issues: Verified deprecation notice for older implementations.
- Retrieve an issue: Current issue attributes and evidence fields.
- List issue events: Occurrence sampling and full-body page limits.
- Retrieve an issue event: Individual event details and supported selectors.
- List organization releases: Release inventory and project filtering.
- List release commits: Associated commit evidence and authorization.
- Integration webhooks: Raw-body signing, response deadlines, and testing guidance.
- Issue webhooks: Lifecycle actions and supported creation categories.
- Update an issue: Writable attributes and elevated scope requirements.
- AI in Sentry: Native AI features and organization controls.
- Seer: Autofix, investigations, integrations, and the Code Review retirement notice.
- Events and issues catalog, organizations catalog, and releases catalog: Reference navigation used to locate the endpoint-specific documentation.
Start building on Major
Build a Sentry triage app with persistent evidence, an agent for engineering judgment, and workflows that require approval before consequential changes.
About the author
Ritvij Gautam is Head of Growth at Major.
Last reviewed November 19, 2026 by Ritvij Gautam.
Frequently asked questions
- What is Sentry AI?
- Sentry AI includes Seer, its debugging agent, plus features such as issue summaries, Query Assistant, AI issue detection, and feedback and replay summaries. Major adds governed cross-system execution with deterministic apps that hold state, agents that exercise judgment, and workflows that manage approvals.
- What does Sentry Seer do?
- Seer investigates errors and performance issues using telemetry and connected code. Autofix performs root cause analysis and proposes fixes, with PR creation available when configured. Seer Agent supports conversational investigation and actions. The docs mark Code Review for removal on December 8, 2026.
- Does Major have a native Sentry connector?
- No. Sentry connects through Major's Pipedream connector, catalog app p_sentry. IT configures the connection once and centrally scopes credentials. Do not confuse this connection with Seer or with a native Sentry connector.
- Which Sentry API permissions are needed for triage?
- Use event:read for issue and event reads, org:read for listing organization projects, and project:read for the release reads described here. Updating an issue requires event:write or event:admin. Start read-only and check each endpoint's documented scopes before granting writes.
- Can Sentry automation resolve issues without approval?
- The API supports issue status updates, but the design here requires human approval before resolution, archiving, or priority reductions. The agent prepares the recommendation, the app stores the evidence, and the workflow applies the exact approved change after checking current state.
- How do Sentry webhooks handle verification and failures?
- Verify Sentry-Hook-Signature using HMAC-SHA256 over the exact raw request bytes with the integration Client Secret. Sentry asks receivers to respond within one second. The reviewed webhook reference does not document a retry schedule, so keep durable intake, deduplication, and reconciliation in your design.