Devin Oncall
Product · SaaS / cloud software by Codeium
Autonomous 24/7 incident response and bug triage agent for engineering teams.
Devin Oncall is an autonomous 24/7 incident response agent that monitors Slack to triage bugs and investigate production issues. It uses long-term memory and observability integrations to diagnose root causes and route issues to the right engineers.
- Languages
- English
Category: AI Agents
- Deployment
- Cloud-based automation
- Limits
- Per-session spend limits and rate limits
- Monitoring Frequency
- 24/7
- Primary Interface
- Slack
- Observability Protocol
- Model Context Protocol (MCP)
- Memory Type
- Persistent scratchpad
- 24/7 autonomous monitoring of Slack channels for bugs and incidents
- Automated root cause analysis by tracing code and querying logs via MCP
- Intelligent deduplication and routing to correct code owners
- Persistent long-term memory via a shared scratchpad
Overview
Devin Oncall is an autonomous incident response and bug triage agent designed for engineering teams to manage production issues 24/7. Operating as a persistent automation, it monitors Slack channels to detect, investigate, and diagnose incoming bugs, regressions, and incidents without manual intervention. The agent utilizes a parent-child architecture where a long-running monitor filters noise and spawns sub-sessions to perform deep-dive investigations into the codebase. By integrating with observability platforms like Datadog and Sentry through the Model Context Protocol (MCP), Devin Oncall correlates alerts with runtime logs and metrics to identify root causes. It maintains long-term memory via a shared scratchpad, allowing it to deduplicate reports and route actionable issues to the correct code owners.
Incident Response and Triage
- 24/7 Slack Monitoring: Devin Oncall monitors Slack channels 24/7 to automatically investigate bugs, regressions, and incidents as they arrive.
- Noise Filtering and Deduplication: A long-running parent Devin listens to every new message, filters out noise, and detects duplicates.
- Automated Root Cause Analysis: The agent spawns focused child sub-devins to investigate actionable bugs by reading relevant code and tracing root causes.
- Incident Diagnosis and Routing: Devin posts a diagnosis in the Slack thread and tags the correct code owner based on its findings.
Knowledge and Memory
- System Learning: Devin Oncall learns the user's systems during setup to provide more accurate investigations.
- Long-term Memory: A persistent scratchpad allows the agent to maintain context, record duplicate items, and store information that persists across session restarts.
- Shared Context: The parent monitor and child sub-devins share the scratchpad to track triaged items and maintain routing tables mapping code areas to owners.
Integrations and Observability
- Datadog Integration: Devin connects to Datadog via MCP to pull metrics, logs, and traces to correlate issues with runtime behavior.
- Sentry Integration: The agent uses the Sentry MCP to look up error details, stack traces, and affected users.
- Linear Integration: Native Linear tools allow Devin to check for related tickets or create new ones without needing a separate MCP.
- Plugin Marketplace and MCPs: Users can enable MCP servers or install plugins from the marketplace to give the agent access to external tools.
Control and Customization
- Customizable Setup Prompts: The setup prompt allows users to customize behavior, such as focusing on specific services or requiring error logs before investigation.
- Spend and Rate Limits: Users can set per-session spend limits and rate limits to control resource usage for the automation.
- Network Access Control: Network policies can be enabled to restrict outbound access for the auto-triage automation when handling untrusted user input.
Use cases
- Engineering teams monitor Slack bug channels 24/7 to automatically filter noise and identify actionable regressions.
- DevOps engineers integrate Datadog and Sentry to allow Devin to correlate production alerts with specific code changes and logs.
- Incident responders use Devin to automatically deduplicate repeated reports and tag the correct code owner for a fix.
- Teams maintain a persistent routing table in Devin's scratchpad to ensure bugs are always directed to the appropriate service owners.
Specifications
- Deployment
- Cloud-based automation
- Integrations
- Slack, Datadog, Sentry, Linear, GitHub, GitLab, Bitbucket, Microsoft Teams, Jira, PagerDuty, Pylon
- Limits
- Per-session spend limits and rate limits
- Monitoring Frequency
- 24/7
- Primary Interface
- Slack
- Observability Protocol
- Model Context Protocol (MCP)
- Memory Type
- Persistent scratchpad
- Integration Support
- Datadog, Sentry, Linear, GitHub, GitLab, Bitbucket
- Resource Controls
- Per-session spend and rate limits
How it works
- Connect Slack Channel: Invite Devin to the specific Slack channel (e.g., #bugs or #incidents) and connect your personal Slack account in settings.
- Configure Automation Template: Create a new automation using the Triage Bug Reports template and select the monitored channel.
- Customize Behavior Prompts: Provide instructions in the setup prompt to define how Devin should handle specific bugs, such as tagging certain teams for frontend issues.
- Connect Observability Tools: Enable MCP integrations for tools like Datadog or Sentry to give Devin access to runtime metrics and error logs.
FAQ
How does Devin Oncall handle incoming bug reports?
Devin Oncall uses a long-running parent Devin to monitor Slack channels, filter noise, and detect duplicates. When an actionable bug is found, it spawns child sub-devins to read code, trace root causes, and post diagnoses.
How does the agent maintain context over time?
The agent uses a persistent scratchpad as long-term memory to track triaged items, maintain routing tables, and record duplicate reports. This context persists across session restarts and is shared between the parent monitor and child sub-sessions.
What monitoring tools does Devin Oncall support?
Devin Oncall integrates with observability tools like Datadog and Sentry via the Model Context Protocol (MCP) to access runtime data, logs, and stack traces. It also has native integration with Linear for ticket management.
Can I customize how Devin triages issues?
Users can customize behavior through a setup prompt, where they can instruct the agent to focus on specific services, require certain data like stack traces, or specify how to tag teams.
Added by AntSpace from public sources (docs.devin.ai).