Polaris
The Tempered Path
Scores the call out of 100 against the eight fixed dimensions, quoting the transcript for every point won and lost.
Stormbright by Dustorm PreSales reads a demo-call transcript and hands the Solution Consultant four things before the follow-up is booked: a grade with the evidence quoted, the full script for the next call, a run-of-show to rehearse from, and a card they can glance at mid-demo.
A workspace is one team’s shared space: its people, its calls, its business context and the rubric everyone is graded against. The tour walks a new one from signup to its first grade on a sample call, with the real agents doing the work. Nothing to install.
Evidence · 00:14:20 · “Let me just walk you through the sales module first, and then we can talk about your close process.”
No new meeting to schedule, no reviewer to find, no coaching queue. The material for the next demo comes out of the last one.
Stormbright reads what you sell, your differentiators and your point of view into a business context library your administrators can edit. Every agent is grounded in it, so nothing on the page is generic advice about demos.
Less than a minute, then ten documents you can open and correct.
Plain text, Markdown, WebVTT, SRT or a transcription API export. Names, companies, products and competitors are substituted for fixed labels before the transcript reaches a model, and you see the substitution table.
One call. No recording integration to install first.
Watch each agent land in turn: the grade arrives first and is readable while the script is still being written. Then edit the cast, re-run a single agent, export what you need and take it into the next call.
A couple of minutes, dominated by Helios writing the long script.
PreSales is a light in a dark world: the buyer needs the information, and they would rather have it from someone who is there to explain than from someone who is there to sell. Every agent here is named for a way of carrying light, and every one does a job you could name without the metaphor. Five of them run on your call. Nexus is the sixth, and it is an administrator's tool that never touches one.
Swaps every company, customer, product and speaker name for a fixed label before a model sees a word of it. In the product this is a substitution table run in code, not a model call, which is why it is a step in front of the chain rather than one of the five.
The Tempered Path
Scores the call out of 100 against the eight fixed dimensions, quoting the transcript for every point won and lost.
The Living Spark
Casts two to four believable humans for the story the next demo will tell, tied to the process the buyer described.
The Forging Core
Writes the next call in full, word by word, built to score Exemplary on every dimension. Read it as a story, not a memory test.
The Guiding Flame
Condenses the script into a rehearsal table: summary, cues, screen. By deletion, never by paraphrase.
The Focused Ember
One column of short lines in the script's own words, for a presenter who already knows the story and needs one glance.
Watches the prompts, the rubric and the business context for drift, and reports what a change would cascade into.
A grade nobody acts on is a number. Every run ends in something a Solutions Consultant opens at a specific point between this call and the next.
A score out of 100 across the eight dimensions, with the transcript quoted for every judgement and a rewritten line where a line went wrong.
Read it the same afternoon
The whole next call, written out. Fixed sections with a named cast, a challenge, the actions and their results, the discovery question, the transition.
Study it the night before
One table per section: what you are saying, the delivery and screen cues lifted verbatim, and the slide they belong to. Enough to rehearse from without reading prose.
Run it in the practice session
A single column of short lines in the script's own words. No cues, no stage directions, nothing to lose the room over. One look and you are back on the beat.
Keep it open during the demo
Every document exports as HTML, Markdown, plain text or JSON, and prints to PDF. Aurora’s cast is not on this list on purpose: it is editable before Helios runs, then embedded verbatim at the top of the script.
A demo rubric that changes under you is a rubric nobody trusts. The eight dimensions and their weights are constants in the code: they cannot be renamed, reordered, removed or added to. What your administrators do own is the language, the band descriptors and the clarifications your team is actually graded against.
Highlighting Differentiators is scored against your differentiators, from your business context library, so the fifth dimension reads with your company’s name in it.
Recordings of customer calls are the most sensitive material a PreSales team owns. Six commitments, each one a thing this system does rather than a policy it has.
Polaris writes the evidence next to the judgement, at the timestamp, in the presenter's own words. It grades from the transcript text alone: never the screen it cannot see, never the tone it cannot hear. Because the quoted line sits next to the number, anyone can check one against the other, and a score a team can check is a score a team can act on.
Every agent's system prompt, template, model and ceiling is editable by an administrator. Editing one bumps its version and keeps the revision it replaced, so you can read what changed, side by side, and put an earlier version back. Every artifact records the prompt version that produced it.
The sixth agent, and the one nobody demos. It reads the prompts, the rubric and the business context against the artifacts they produced, reports where they have fallen out of alignment, and tells you what a proposed change would cascade into before you make it.
Isolation is enforced in the database with row-level security rather than in application code, and it is tested as such. Roles are real: Solutions Consultant, Manager, Administrator, each seeing a different product.
Before a transcript goes to a model, Stormbright can replace the company, customer, product, competitor and speaker names you list with fixed labels. It is deterministic text replacement rather than a model, so running it sends nothing anywhere, and it reports the sensitive-looking terms it was not told about. It is off until you turn it on, and the original transcript stays in your workspace either way. The security page sets out exactly what it does and does not catch.
Stormbright runs on Google Cloud in us-central1. Vertex AI is the only place a model is called from, which keeps the question of where your transcripts go to a single, answerable one.
The detail behind these, including where transcripts are stored, who can read them, and Google’s own published commitments for Vertex AI, is on the security page.
Bring one transcript from a demo that went fine but not brilliantly. That is the one Stormbright has the most to say about.