Oko
Oko/Docs

Command reference

oko telemetry inputs

Which inputs take the sessions' context and tokens. Each OMP call records the context it was sent with; the growth between two calls is shared among the entries appended between them by their characters, and each entry stays in the context, costing its tokens again on every later call, until the next compaction. So the inputs (the operator's messages, the model's answers, each tool's results, each hook's paragraph of a stop continuation, compactions, the system prompt) add up to the context the calls sent.

Invocation#

Command
oko telemetry inputs --since WINDOW [--session ID ...] [--limit N] [--root DIR ...]

Inputs and options#

  • --since, required: 7d, 48h, a day as YYYY-MM-DD or an RFC 3339 time.
  • --session, repeatable: only those sessions.
  • --root, repeatable: only those trees; every harness's own transcript tree otherwise. --limit caps the transcripts read.

Output and state effect#

  • Reads only; prints `{since, roots, files, contextTokens, unreadable, notMeasured, inputs, sessions}`. Each input row carries `entries`, `chars`, `tokensAdded`, `tokensCarried`, `shareOfContext`, `callsAfter`, `secondsAfter` (and `sessions` in the totals). Each session carries its calls, context windows, context and output tokens, context per call, seconds in calls, models, inputs and `repeats`: every tool call made again with the same arguments, how often and what its results took.
  • A transcript that is not OMP's, or whose name names no session, is listed under `notMeasured`; a line that is not JSON under `unreadable`. Oko Desktop's Activity › Context inputs and `oko sessions scorecard` (each card's `tokens`) show the same measurement; the native API answers it as `stats.inputs`.

Refusals and safe exits#

  • `oko telemetry inputs requires --since <text>`; `oko telemetry inputs --since takes a window like 7d or 48h, a day as YYYY-MM-DD or an RFC 3339 time, not <value>`; `oko telemetry inputs: --limit takes a whole number, got <value>`.

CLI reference#