Skip to content
runmarq

Know when the work finishes.

Build from sourceSee how it worksGitHub
  • Claude Code
  • Codex
  • Cursor
Demo

Claude Code · acme-web

Add password reset to the login flow

Running
Elapsed time
00:00
Estimate
About 7–9 minA few minutes
Prediction confidence
Medium

The agent accepts the prompt. runmarq classifies it locally and makes a first estimate.

Starting · 0 tool calls

0:00 Prompt submitted

Know what to expect.

A likely range with a plain answer beside it: almost done, a few minutes, or time for something else. While the evidence is thin, the plain answer leads. The range tightens as the agent works, says why when it moves, and switches to “Finishing up” once your own runs show the work is wrapping up.

Estimate

About 6–11 min

Time for something else

Elapsed
4 min
Most likely
About 9 min more
Prediction confidence
Medium

Work without checking.

The estimate lives in the tray. A quiet notification arrives when a run finishes, fails, needs your approval, or looks stuck.

Claude Code finished

Run in acme-web · 7 min

See where the time went.

Every run keeps its phases, waiting time and the estimates made along the way, so you can see what slows your agents down.

Phases

Run finished · 7 min

  • Reading files56 s · 6 calls
  • Editing code1 min · 4 calls
  • Running tests3 min · 3 calls · 2 failed
  • Waiting for you42 s
  • Editing code38 s · 2 calls
  • Running tests34 s · 1 call

Sharper with every run.

Four models learn from the runs on this computer and, once enough have finished, count by how well each predicted your recent ones. None of it spends your agent's model usage.

Estimation model

  • History baseline64 runsIn use
  • Phase simulation61 runsIn use
  • Similar prompts64 runsIn use
  • Learned model64 of 100 runsNot in use

The numbers, not the text.

Token counts, context size and working-tree changes are read from the agent's transcript and from git on this computer. They sharpen the estimate; the text itself is never kept.

Run activity

Effort
High
Phase
Editing
Model responses
35 + 6 by subagents
Output tokens
16.4k (2.1k thinking)
Context in use
84k of 200k
Compactions
1
Lines edited
118
Working tree
4 files, 132 lines changed

In your terminal too.

Chain your next step onto the agent finishing with runmarq wait, and keep every run in your prompt or Claude Code's status line.

$ runmarq wait && npm run deploy

Waiting for Claude Code · acme-web · 4:12 elapsed · A few minutes, about 2–5 min

✓ Claude Code · acme-web finished after 6:12.

$ runmarq statusline

runmarq ● 2 working · next: about 3–7 min

Every feature

Connect once. Keep working.

How it works

  1. Claude Code

    Detected · 2.1.143
    Approve and connect

    Connect your coding agent.

  2. Claude Code · acme-web

    Feature

    Running tests · 03:14

    Estimate

    A few minutes

    About 4–9 min

    Keep working as usual.

  3. Check the estimate when you need it.

This computer

  1. Claude Code, Codex, Cursor

    official hooks

    then

  2. runmarq emit

    normalize, minimize, spool if offline

    then

  3. Transcripts, git

    counts only, no text

    then

  4. Collector

    run state, estimator, SQLite history, local API

    then

  5. Desktop app, tray, notifications

Three agents. What each one reports.

  • Prompt start
  • Turn completion
  • Tool start
  • Tool completion
  • Plan progress
  • Approval requests
  • Errors
  • Subagents
  • Model metadata
  • Token counts
  • Interrupts reported directly
  • Claude Code does not report user interrupts directly. runmarq infers them and marks those runs cancelled, never finished.
  • Plan progress is available only when Claude keeps a task list or to-do list.
  • Model metadata is reported at session start, not for every prompt.
Claude Code setup

Your code stays yours.

Privacy statement

Kept on this computer

  • When runs start, wait and end
  • Tool names and activity types
  • A task category from a local classifier
  • Token counts and change sizes, without the text
  • A salted repository identifier and its folder name, which you can turn off
  • Forecasts and how they compared with reality

Never stored or sent

  • Prompt text, unless you turn on short titles
  • Source code and file contents
  • Command arguments and output
  • Transcript text and the names of changed files
  • Environment variables, API keys and secrets
  • Anything to a server: there is no account and no analytics

Spend less time watching the terminal.

Build from sourceRead the docs

Claude Code · acme-web

Feature

Run finished · estimated 4–12 min

Actual duration

8 min