The Ember Codex
The one-change report: a test plan for your four free Ember Console reports
Ember Console's free plan runs four reports a month. Run the first as written, then put exactly one change into each of the next three.
Ember Console’s free plan runs four reports a month, and the product is in closed beta. Four is enough for a test if you spend them in order: run the first as written and read it, then put exactly one change into each of the next three. I call that a one-change report, because afterwards you can name which change did the work.
The pricing page states the allowance plainly: “Four reports a month, free for as long as you like. No card, no countdown.”
What goes into each of the four reports?
Report one is the reading. You run the template you already have and change nothing, because you do not yet know which box is weak.
- Report one: run the template as written. Change nothing.
- Report two: tighten one prompt. Same template, same sections.
- Report three: sharpen one eval. Leave every prompt alone.
- Report four: edit in the dashboard. Touch no prompt, no eval.
In Ember Console a template is your letterhead plus a set of boxes, each box holding a prompt and an eval. Report two leans on the prompt, report three on the standard the box has to pass, which is the choice a wrong box forces on you.
Report four pays late. Ember builds your next report with your earlier edits in mind, so a report-four edit shows up in the report after it, which may be next month’s first.
Why only one change per report?
Because a prompt and an eval fail differently, and a report that improves after you change both cannot say which was broken. A vague prompt gives you a thin box. A loose eval lets the thin box through. On the page both look the same, and only the order of the test separates them.
Three things stay still: the same kind of session, the same template, and the same AI. Ember is reachable from ChatGPT, Claude or Littlebird, so a model swap halfway through costs you a report.
The published test advice answers a question you have already settled. Simular tested eight note takers across 50-plus meetings, and AssemblyAI’s 2026 guide says to “test 2-3 against your real meetings before paying”. Both pick the tool. In a closed beta you have the tool; what is untested is the template, the prompt and the standard a box has to pass.
What is not stated yet?
Ember Console counts four reports a month. What a report counts as, I have not said.
Whether the count is one recording, one built report, or one rebuild of a report you already sent. Whether last month’s recording counts. And what the month runs from, so you know when the fifth arrives.
Until I write those down, run the four inside one calendar month and ask me if you need a fifth.
Pick the change for report two now, while report one is still fresh enough to argue with.
Questions people ask next?
How do you test an AI note taker?
Run it on a session you have already written up by hand and read the two side by side. Then change one thing per report: the prompt, the eval, the edit, in that order.
Is there a free AI note taker?
Ember Console’s free plan costs $0, needs no card, and runs four reports a month with “API & MCP, connects to any AI”. Every plan keeps your letterhead and exports to PDF or Word.
What happens after the fourth report?
Individual costs $8 a month billed yearly, or $12 month to month, with unlimited reports and templates. The free four are monthly, but what the month runs from is not stated, so do not schedule around a fifth.
Ember Console is a session-report builder for coaches, in closed beta at $8 a month billed yearly with 4 free reports a month: https://ember.anyeads.com/. There is nothing to sign up for yet.
Ember