[gkarthi-signoz]

Summary

Adds a dashboard template for eve, Vercel's filesystem-first agent framework. It reads the OpenTelemetry traces eve emits through otelIntegration() from eve/instrumentation/otel. Traces only; eve emits no metrics or logs by default.

• eve/eve-dashboard.json: v6 dashboard, 23 panels in 5 sections
• eve/README.md: one section per panel
• eve/images/: per-panel screenshots

Telemetry shape

Measured against eve 0.66.0 in interactive eve dev mode, scope eve.agent:

invoke_agent <agent>          root, one trace per turn; agent.turn.outcome
└─ agent.step                 one per model step
   ├─ chat <model>            gen_ai.* request, response, usage
   └─ agent.action            agent.action.kind = tool-call | subagent-call
      └─ execute_tool <tool>  absent for provider-executed tools such as web_search

Subagent turns are their own root traces, linked to the caller, sharing gen_ai.conversation.id.

Load-bearing query choices

Choice	Why
Token panels sum chat spans only	eve reports identical usage on chat, agent.step, and invoke_agent; summing all spans triples it
Failure panels use agent.turn.outcome = 'failed'	A cancelled turn leaves invoke_agent UNSET but marks its children ERROR TurnCancelledError
LLM Error Rate and Recent Errors exclude TurnCancelledError	Otherwise every Escape in the TUI counts as a model failure
Actions by Kind reads agent.action	Only way to count provider-executed tools and subagent calls
Tool panels read execute_tool	Tool failures never fail the turn

Panels

Section	Panels
Overview	Turns, Conversations, LLM Calls, Input Tokens, Output Tokens, Turn Failure Rate
Tokens & Models	Token Usage Over Time, LLM Calls by Model, Tokens by Model, Tokens by Agent
Latency	LLM Call Latency, Turn Latency (p95), Model Steps per Turn, Finish Reasons
Agents, Tools & Subagents	Agents, Tools, Tool Calls Over Time, Actions by Kind
Outcomes & Errors	Turn Outcomes, Errors by Type, Tool Error Rate, LLM Error Rate, Recent Errors

Notes

• The README links https://signoz.io/docs/eve-observability/, which does not resolve yet. The guide is in progress in signoz.io.
• Screenshots come from a week of data across three services, shaped on real eve traces.

[github-actions]

📊 Dashboard files added in this PR

• eve/eve-dashboard.json

✅ An issue has been created in SigNoz/signoz.io to track these dashboard changes.