Sevbot

OpenAI's internal incident-response agent, built on Codex. When an incident is detected, the bot wakes up, collects context, determines possible mitigations without executing any, and answers developers' questions in the incident Slack channel; an engineer can tell it to apply a chosen mitigation.

Company: OpenAI
Approach type: Agent
Work: On-call
Human involvement: Drafts reviewed
Invocation: Event driven
Interfaces: Slack
Deployment stage: Deployed
Evidence strength: Secondary only
Entry reviewed: 2026-09-16

How it works

The workflow the sources report for this implementation.

The bot determines possible mitigations but never executes any; an engineer tells it to apply a specific mitigation

Where people stay involved

Each scope pairs its normal attention boundary with supporting evidence. See thesupervision definitions for the level mapping and limits.

incident detected -> applied mitigation
Work product review · Level 3

Catalog interpretation

Level 3 for incident detected -> applied mitigation; human attention boundary: work-product-review.

Observed in September 2026

Implementation details

Model: unknown
Harness: Built on top of Codex
Sandbox: unknown
Tool access: N/A
Knowledge: N/A
Context management: N/A
Credentials: N/A
Interfaces: slack

Reported results and limitations

The catalog records what the sources report, with the scope and the denominator of every figure. A qualification below limits the figure it sits under.

Lessons and interpretation

OpenAI's stated goal is that Sevbot mitigates routine outages autonomously with humans reviewing its actions on return; as of the report, on-call duty remains

Observed in September 2026

Sources and research details

Citations link to the original publisher. Each source also keeps a preserved copy in the repository, so a changed or removed page stays checkable.

Inside OpenAI's agentic software factoryhttps://newsletter.pragmaticengineer.com/p/openai-software-factoryNews · Independent secondary · Last source verification: 2026-09-16
Preserved copy
Certificate Transparency records for sevbot.api.openai.comhttps://crt.sh/?q=sevbot.api.openai.comOther · Aggregator · Last source verification: 2026-09-16
Preserved copy
[SEVBot] Optionize initialized notification tolerance (openai-agents-python pull 2765 commits)https://github.com/openai/openai-agents-python/pull/2765/commitsSource code · First party · Last source verification: 2026-09-16
Preserved copy

Research details for every claim on this page

Summary
Statement type
Fact
Provenance
Reported
Confidence
High
Confidence reason
A linked first-party source states the claim.
- SupportsInside OpenAI's agentic software factoryPreserved content.md, lines 188-193 (pipeline step 9: respond to outages)
- SupportsCertificate Transparency records for sevbot.api.openai.comPreserved content.md, lines 13-41 (certificate table for sevbot.api.openai.com; earliest not-before 2025-11-20 at lines 37-41, with renewals through 2026-08-01; evidences a deployed OpenAI service under the Sevbot name, not its incident-response behavior)
- Supports[SEVBot] Optionize initialized notification tolerance (openai-agents-python pull 2765 commits)Preserved content.md, lines 21-27 (initial commit '[SEVBot] Optionize initialized notification tolerance', 219a67a, 2026-03-24, authored by an OpenAI employee and co-authored by Codex; evidences internal Sevbot-branded engineering work)
Sandbox
Statement type
Inference
Provenance
Catalog judgment
Confidence
Medium
Confidence reason
The article does not document an execution sandbox; unknown does not mean absent.
- SupportsInside OpenAI's agentic software factoryPreserved content.md, lines 188-195 describe Sevbot without documenting an execution sandbox
Harness
Statement type
Fact
Provenance
Reported
Confidence
Medium
Confidence reason
A linked participant or independent source reports the claim.
- SupportsInside OpenAI's agentic software factoryPreserved content.md, line 188 (Sevbot is built on top of Codex)
Model
Statement type
Inference
Provenance
Catalog judgment
Confidence
Medium
Confidence reason
The article names no underlying models; unknown does not mean absent.
- SupportsInside OpenAI's agentic software factoryThe preserved article names no underlying models for Sevbot
Interfaces
Statement type
Fact
Provenance
Reported
Confidence
Medium
Confidence reason
A linked participant or independent source reports the claim.
- SupportsInside OpenAI's agentic software factoryPreserved content.md, line 192 (it is part of the Slack channel and answers devs' questions)
Supporting component
Statement type
Fact
Provenance
Reported
Confidence
Medium
Confidence reason
A linked participant or independent source reports the claim.
- SupportsInside OpenAI's agentic software factoryPreserved content.md, lines 190-193 (mitigation bullet list)
Lesson
Statement type
Fact
Provenance
Reported
Confidence
High
Confidence reason
The stated goal and the current on-call state are reported directly.
Observation date
2026-09
- SupportsInside OpenAI's agentic software factoryPreserved content.md, line 195 (goal of autonomous action on routine outages; oncall duty is not a thing of the past)
Operating model assessment
Statement type
Inference
Provenance
Catalog judgment
Confidence
High
Confidence reason
The article enumerates the boundary: Sevbot proposes mitigations and never executes any; an engineer commands the application.
Observation date
2026-09
- SupportsInside OpenAI's agentic software factoryPreserved content.md, lines 190-193 (determines possible mitigations but never executes any; an engineer can tell it to apply a specific mitigation)