Operational investigations that show their evidence

Operational AI should not jump from a symptom to a confident root-cause story. Opsphere now structures investigations around multiple possible explanations and progressively tests them against what the connected operational systems actually report.

Hypotheses can cover deployments, configuration, infrastructure, application behavior, external dependencies, networking, CDN behavior and other operational domains. Evidence collected from real tool calls can support, reject or leave a hypothesis unresolved, while observed facts remain clearly separated from inference.

Confidence is calculated from the available evidence, including its quality, independent supporting sources and contradictions. When the underlying systems provide timestamps, Opsphere also builds an investigation timeline from the evidence already collected, without making unnecessary additional requests.

Supported conclusions and recommendations can include explicit verification conditions: what should become true before an issue or proposed resolution is considered validated. Opsphere remains read-only by default — it investigates, correlates and recommends without silently changing production. High-level investigations and macros on the MCP gateway can return the same structured result to Cursor, Codex and other MCP-compatible clients.