The failure starts the investigation.
spanshot begins from failure signals as they happen, instead of waiting for someone to assemble the evidence.
Investigate failures from the telemetry your system already emits.
spanshot turns OpenTelemetry signals into focused, inspectable investigations for AI agents and developers.
Failures leave signals. Investigations still start from scratch.
Distributed systems continuously emit traces, logs and events. But that evidence is fragmented across tools. Before they can reason about a failure, AI agents and engineers still have to find, correlate and reconstruct the relevant context.
The investigation is the unit of work.
- API gateway
- Checkout service
- Payments service
- Queue worker
- Database
spanshot
Investigationfocused, inspectable evidence
Dashboardinspect evidence
- MCPContinue in your agentClaude Code · Codex
- Async harnessRun in your automationHermes · OpenClaw
- Managed spanshot harnessLet spanshot run the investigationspanshot-operated
How the investigation continues is your choice
The investigation is the same. These are alternatives, not steps. Pick one, or use them side by side.
You ask, in your own tooling
Ask from your agent
The investigation is exposed over MCP, so the agent you already work in can pull it in and reason on it.
- MCP toolget_investigation
A failure arrives, nobody is watching
Start automatically
When an investigation is created, a webhook hands it to your existing agent harness with the evidence attached.
- Webhook eventinvestigation.created
You want the analysis already done
Let spanshot investigate
spanshot runs the analysis with your own model key and delivers the result to a channel you configure.
- Managed deliveryYour configured channel
Principles behind every investigation
Ready for an AI agent or an engineer to investigate.
The behavior relevant to a failure is organized before the analysis begins.
See what the analysis is based on.
The original telemetry stays available alongside the investigation, you inspect the evidence, not just an AI conclusion.
What spanshot is not
Not error monitoring
Error monitoring tells you that something broke, how often, and where it was thrown.
spanshot organizes the behavior around one failure across services, so it can actually be reasoned about.
Not a general coding agent
An agent working from the repository infers what probably happened from the code it can read.
spanshot hands it the runtime evidence of what did happen, from the telemetry the system emitted.
Not a dashboard with a chat box
Asking questions over charts still leaves you assembling context and trusting a summary.
The original telemetry stays attached to a focused investigation. The system is open source and OpenTelemetry-native, with no proprietary instrumentation.
Start the investigation from the telemetry you already have.
spanshot prepares a focused, inspectable investigation for developers and AI agents.