Pipeline Monitor
Real-time message flow visualization
The Pipeline Monitor provides a real-time, visual trace of every message flowing through GNETiX. It shows the complete journey from message receipt through LLM reasoning, tool calls, and response delivery as an interactive directed graph.
What It Shows
Every message processed by GNETiX generates a trace -- a sequence of events that represent each step in the pipeline. The Pipeline Monitor renders these traces as a directed graph with distinct node types:
| Node Type | Description |
|---|---|
| Message Received | The incoming message from Webex, Slack, Teams, or a webhook |
| Director | The orchestration engine receiving the message and building context |
| LLM Iteration | Each call to the LLM (there may be multiple if tool calls trigger follow-up reasoning) |
| Reasoning / Thinking | The LLM's chain-of-thought or extended thinking output |
| Tool Call | A tool invocation dispatched to an MCP server via the on-prem agent |
| Message Sent | The final response delivered back to the user on their platform |
Real-Time Streaming
Trace events are delivered to the portal via Server-Sent Events (SSE). As the Director processes a message, each step appears in the graph in real time -- you can watch the AI think, call tools, and compose its response live.
Trace Teaser Cards
The Pipeline Monitor shows a scrollable list of recent traces as teaser cards. Each card displays:
- Sender name and platform
- Message preview (first line of the user's message)
- Timestamp
- Token usage (input tokens, output tokens)
- LLM cost (dollar amount for the trace)
- Duration (total processing time)
Click any teaser card to expand the full trace graph.
LLM Spend Tracking
Every trace records the token counts and cost for each LLM call within it. This gives you per-message cost visibility:
- Input tokens and output tokens per LLM iteration
- Cumulative cost for the entire trace
- Model used for each iteration (the LLM Router may select different models for different tasks)
Cost tracking requires LiteLLM to be configured with pricing data for your models. GNETiX reads cost information from the LiteLLM response metadata.
Filtering
Use the filter controls to narrow the trace list:
- By sender -- see traces for a specific user
- By platform -- filter by Webex, Slack, Teams, or webhook
- By date range -- view historical traces
API Access
Traces are also available via the REST API:
| Method | Endpoint | Description |
|---|---|---|
GET | /api/traces/recent | List recent traces |
GET | /api/traces/{trace_id} | Get a specific trace with all events |
GET | /api/traces/stream | SSE stream of trace events in real time |