See every coding agent before context or limits bite.

A browser-based operations model for Claude Code, Codex CLI, and OpenCode sessions. Add sessions manually or simulate a fleet to learn the monitoring signals that matter.

Active sessions0

Tokens today0

Avg context0%

Rate pressure0%

Orphan ports0

Session table

0 visible

How to use this model: context pressure predicts compaction or degraded recall; rate pressure predicts throttling; long runtime can indicate stuck work; and a listening port without a matching live session may be orphaned. In a real tool, discovery should run locally and expose only the minimum telemetry users approve.

Understanding Coding Agent Session Health and Telemetry Models

How do context window limits, rate-limit consumption, and orphan ports determine coding agent health status in this simulation?

This interactive dashboard demonstrates how autonomous coding agent operations can be monitored across parallel sessions. The simulation evaluates each session into one of three statuses: critical, warning, or healthy. A session is marked critical if context or rate pressure reaches 95%, or if an orphan listening port is detected. A warning is triggered if context or rate pressure meets or exceeds the user-configured thresholds (initially 75% and 70%) or if elapsed runtime exceeds 120 minutes. All underlying values advance via synthetic in-browser timers rather than live daemon socket polling or vendor API hooks.

The monitor operates purely as an in-browser mathematical toy. Telemetry values (tokens, runtime, context percentage, and rate pressure) increment via randomized interval steps (+520 tokens/sec max, +0.05 min/sec runtime) rather than live inspection tools like lsof, netstat, or LLM provider token billing counters. Real production monitoring requires host-level daemon permissions and authenticated API access.

Try a worked example

Click 'Pause simulation' to halt the synthetic metric increments, then locate the 'Context warning threshold' slider. At the default 75% threshold, the mobile-release session (context at ~81%) displays an amber warning bar. When the slider is adjusted or filtered by status, sessions dynamically reclassify: selecting 'Critical' in the status filter isolates only sessions that have breached 95% pressure or harbor an orphan port.

Health Classification Logic

The client-side health function evaluates each session object through strict tier checks. If the orphan port flag is true, or context utilization reaches 95% or greater, or rate pressure reaches 95% or greater, the health is assigned 'critical'. If it escapes critical criteria, the session is assigned 'warning' if context percentage is greater than or equal to the configurable context threshold, rate pressure is greater than or equal to the rate threshold, or runtime exceeds 120 minutes. All other sessions evaluate as 'healthy'.

Browser Boundary and Port Scanning Simulation

Web browsers cannot directly inspect local network sockets or host process tables due to sandboxing boundaries. Clicking 'Scan orphan ports' in this tool filters the in-memory array for objects where the orphan property is already set to true, appending a log message and re-rendering rather than issuing system calls or OS-level process queries.

Simulated sessions and unfiltered aggregate metrics

Your browser does not support video.

Read the explanation

The source seeds four sessions. Docs refresh is paused; the other three are not. Active metric counts unpaused entries, so it shows three even while the table has four rows. Bars compare these counts. The dashboard is a simulator: it does not discover actual Codex or other agent processes, inspect their ports or connect to real telemetry. The token metric sums all four sessions, including paused docs refresh, to one hundred eighty thousand nine hundred. Filtering table rows does not narrow aggregate metrics. The context average is sixty eight point five rounded to sixty nine percent, and rate average fifty four. Bars use point zero zero two four pixels per token to compare total and paused contribution. Simulation ticks can change unpaused values randomly. Health is critical for an orphan, context at least ninety five, or rate at least ninety five. Legacy migration is critical because its orphan flag is true, even though context starts at ninety three. Warning also checks configurable thresholds or runtime greater than one hundred twenty. Paused state changes the dot and activity count but not health classification. Bars compare context percentages. Scan and removal buttons only manipulate the simulated records, not real network or OS processes.