Agents that have been stuck for more than 5 minutes (or 300 simulated seconds) are listed below. You can hand them off to a human.
In this dashboard, each generated run combines prompt tokens and completion tokens to calculate estimated model expenditure. When a new run is added, the tool sums tokens in and out, divides by one thousand, and multiplies by the two-thousandths rate. If an agent run exceeds 300 seconds of inactivity, it enters stuck status until an operator triggers human handoff, restoring it to completed.
How does this monitor evaluate agent status, calculate token costs, and trigger human handoffs?
This dashboard tracks execution runs of autonomous LLM agents by logging step counts, simulated elapsed runtimes, and input/output token usage. The code computes estimated costs using a simplified flat-rate multiplier of $0.002 per 1,000 total tokens (both input and output combined). Agents exhibiting over 300 seconds of inactivity are flagged with a 'stuck' status, exposing them in the Stuck Handoff queue where human intervention marks them completed.
The dashboard applies a single uniform rate of $0.002 per 1,000 tokens across prompt and completion tokens, whereas production LLM providers price input tokens, cached prompt tokens, and completion tokens at distinct rates. In addition, the 300-second stuck threshold relies on client-side timestamp offsets rather than real agent heartbeats or dead-letter queue metrics.
Click the '+ Add Run' button to generate a new simulated run. The tool picks a random status ('running', 'completed', 'stuck', or 'failed') along with randomized token counts, duration, and steps, incrementing 'Total Runs' from 7 to 8 and immediately appending a new 'RUN-008' card to the list.
The client calculates run expenditure as `((tokens_in + tokens_out) / 1000) * 0.002`. This treats prompt tokens and generated completion tokens symmetrically, without accounting for provider-specific tiered pricing or context window cache hits.
A run is evaluated as 'stuck' if `(Date.now() - r.last_activity) / 1000` exceeds 300 seconds. In this simulator, clicking 'Handoff to Human' in the Stuck Handoff tab switches the status to 'completed' and resets the last activity timestamp to the present moment.