Visual Trace Studio

TraceServer serves a browser-based UI for inspecting workflow traces, execution graphs, and RAG pipelines.

Start the studio

meshflow trace-server
# → http://localhost:8765

Or from Python:

from meshflow import TraceServer

server = TraceServer(ledger=ledger, port=8765)
server.start()   # non-blocking daemon thread

Pages

Trace viewer (/)

• Step-by-step execution timeline for any run ID
• Per-step: verdict, cost, tokens, carbon, duration, block reason
• Hash-chain integrity indicator
• Search and filter by node, verdict, or timestamp

Graph view (/graph)

• Visual node-and-edge DAG of the workflow
• Color-coded by verdict (green=approved, red=blocked, yellow=paused)
• Hover for node stats; click to expand step details
• Mermaid and DOT export buttons

RAG builder (/rag)

• Interactive knowledge source configuration
• Test retrieval queries against live VectorStore
• Token budget visualization

Navigation

All three pages share a top navigation bar. The active page is highlighted. Deep-link any page via URL.

CLI trace viewer (text mode)

meshflow trace <run_id>          # terminal trace timeline
meshflow trace <run_id> --json   # JSON output
meshflow trace <run_id> --open   # open browser to trace-server