Reson: AI Incident Commander

Reson is an AI Incident Commander that helps engineering teams investigate and coordinate production incidents through real-time voice conversations.

Instead of acting as a generic chatbot, Reson maintains a structured, shared incident state containing facts, hypotheses, actions, decisions, conflicts, and a timeline. As responders talk, Reson continuously updates this operational picture and uses it to guide the investigation.

The key idea is simple: during an incident, the team should not have to remember everything that was said. Reson turns the conversation into structured operational knowledge.

What Reson Does

• Real-time, interruptible voice interaction with an AI Incident Commander
• Maintains structured incident state throughout the investigation
• Separates facts from hypotheses and actions
• Identifies conflicting evidence and surfaces unresolved conflicts
• Tracks decisions, actions, owners, and incident timelines
• Uses dynamic follow-up questions based on the conversation
• Provides incident-specific reasoning instead of generic chatbot responses
• Supports multiple incidents and allows responders to switch between them
• Generates incident reports for post-incident review

How It Works

A responder joins an incident room and speaks naturally with Reson. Speech is converted to text using Deepgram, processed by an OpenAI-powered reasoning layer, and converted back to speech using MiniMax TTS.

Reson can use MCP tools to read and update the structured incident state. This allows the conversation and the incident workspace to stay synchronized in real time.

The incident state acts as the source of truth:

Facts → Hypotheses → Decisions → Actions → Conflicts → Timeline

For example, during a payment outage, responders might report a spike in payment failures and database connection timeouts. If the database team simultaneously reports that the database is healthy, Reson can recognize the contradictory evidence, record it as a conflict, and guide the team toward the next investigation step.

Why Reson?

Production incidents are noisy. Information arrives from dashboards, logs, engineers, customers, and different teams at different times. Traditional incident tools record what happened, but they do not actively help the team reason through the uncertainty.

Reson sits inside the incident conversation and turns that uncertainty into a continuously updated operational picture.

The goal is not to replace the incident commander.

It is to give the incident commander a second brain that listens, remembers, structures evidence, challenges contradictions, and keeps the investigation moving.

Built With

• Next.js / React
• FastAPI / Python
• Agora Voice AI
• Deepgram STT
• OpenAI LLM
• MiniMax TTS
• MCP
• Cloudflare Tunnel
• TypeScript
• Pydantic

Demo Scenario

The included demo focuses on a Payment Service Outage where payment failures increase shortly after a deployment and database connection timeouts appear.

Reson helps the responder establish the timeline, identify the deployment and connection-pool hypotheses, surface the conflict between application-level timeouts and healthy database checks, and maintain the structured incident state throughout the investigation.

Open-source project with the complete implementation and documentation available in the repository.