PrismaCue
Every card shows where it came from

Your whole workflow builds itself — live and shareable.

PrismaCue is a live, AI-assisted canvas workspace. Context flows in from your org, CRM, live calls, notes, and integrations — and an entire workflow journey assembles in real time: discovery maps, project plans, solution proposals, architecture diagrams. You live-share it with your customer or stakeholder as it builds, and every card shows whether it was said, sourced, or AI-suggested.

~15s
spoken word → cited fact
3-state
Said · Sourced · AI, in code
Verified-only
reaches the customer
Proposed Solution · assembling live capturing
Pain Said

"We lose two days rebuilding every discovery into a solution doc."

Requirement Said

"Everything has to sync back to Salesforce on the Opportunity."

Capability Sourced

Salesforce: artifacts + summaries push out to the Opportunity.

Proposed AI-suggested

Suggested: phased rollout, pilot team first.

Platform credibility

Built to be audited, not just trusted.

Every card's source is labeled in code — not the model's opinion. The same discipline runs end-to-end: isolation probes in CI, an append-only ledger, full cost visibility, and a capture path that never depends on the AI being reachable.

  • Row-level isolation, CI-verified

    Multi-tenant Postgres RLS keeps every deal strictly partitioned. Isolation probes run in CI — not just assumed.

  • Ordered log, full replay

    Every event is appended to an ordered log. The cockpit can reconstruct any session from scratch — no state lost, ever.

  • Token cost on every call

    Per-tenant, per-deal telemetry is captured on every model invocation. You see exactly what each deal costs the model budget.

  • LLM outage? Nothing lost.

    Degraded mode keeps capture running even when the model is unreachable. Facts land in the ledger; generation catches up when the AI returns.

Accuracy is structural, not a policy. Every card's 3-state badge — Said / Sourced / AI-suggested — is derived from deterministic code. No model labels its own source.

How it works →

Built for sales engineering teams

Customer marks pending — not yet in production

The problem

Great discovery call.
Empty deliverable at midnight.

SEs run great discovery calls — then spend nights reconstructing them into solution docs, architecture diagrams, and follow-ups. Every reconstruction loses fidelity. Momentum fades. Deals slip.

Acme — Q3 Discovery · 47:23
call in progress
Buyer

"Our integration team won't touch anything that doesn't have a proper architecture diagram on day one."

Buyer

"We need the recap and a proposed solution doc before the exec review — that's Friday."

Buyer

"The biggest pain is rebuilding tribal knowledge every quarter when engineers rotate."

47 min · speaker-attributed · everything captured
Acme — Proposed Solution.doc
11:47 PM
[ architecture diagram ]
reconstructing from memory…
0 of 47 min cited · grounded in nothing

Fidelity lost

The exact words, constraints, and priorities the buyer stated do not survive manual reconstruction intact. Every paraphrase is a gap.

Momentum bleeds out

Every hour between call end and deliverable is an hour the buyer's confidence erodes and the competition advances.

Nothing builds the doc

Your notes tool transcribes. Your CRM stores. Neither assembles the solution from what was actually said on the call.

PrismaCue closes the gap: the deliverable assembles while the call is still live.

How it works · the prism

Speech enters as light. Cited artifacts come out.

PrismaCue runs discovery through a prism. A spoken word enters as light and exits as a structured, cited solution artifact — and every stage is named for the optics it performs. Nine stages, one pass, no nightly reconstruction.

spoken word Capture Beam · Ray Ground Source · Aperture Generate Refract · Spectrum · Trace Deliver Focus · Wavelength
Beam → Ray → Source · Aperture → Refract → Spectrum · Trace → Focus → Wavelength
  1. Beam

    Ingestion

    01

    Live calls via meeting bot or desktop capture, plus recordings, voice notes, pasted notes, and Salesforce. Every source becomes a Beam with a full speaker-attributed transcript.

  2. Ray

    The atomic captured fact

    02

    Speaker-attributed, verbatim-quoted, and kind-classified — pain, requirement, commitment, ask, metric, question+answer. On live calls, Rays land about 15 seconds after the words are spoken.

  3. Source

    Org knowledge base

    03

    Product docs, collateral, and POVs — the grounding corpus that feeds generation and the Navigator.

  4. Aperture

    Discovery workspace

    04

    Question templates that fill themselves as answers are spoken — the SE's quiet cockpit, six keys or fewer per call.

  5. Refract

    The AI engine

    05

    Per-group fan-out generation, stateless JSON-patch refinement, and groundedness verification. Anthropic models, with a BYO-AI seam per org.

  6. Spectrum

    Solution canvases

    06

    Current State, Proposed Solution, and Proposed Architecture canvases — assembling node-by-node, live, as the conversation unfolds.

  7. Trace

    Diagram generation

    07

    Architecture and flow diagrams drawn on the same canvases, from the same grounded facts.

  8. Focus

    Customer-facing stage

    08

    Branded and presenter-driven. Only verified content ever reaches it — the entailment gate never relaxes.

  9. Wavelength

    Deal-alignment analytics

    09

    Deterministic coverage scoring — which discovery questions are answered, which facts made the solution — rolled up for leadership. A number you can audit.

Through every stage

Each card carries a 3-state badge — Said, Sourced, AI-suggested — derived in code, never by the model. Nothing reaches Focus until it passes the verification gate.

Trust & accuracy

Grounded in what was actually said.

AI hand-waving is easy to sell and impossible to trust. PrismaCue makes accuracy structural, not a promise: every card’s source is labeled in code, verification is a gate that never relaxes, and coverage is a number you can audit. Nothing reaches your customer on the model’s word alone.

Three states. Derived in code, never by the model.

Every card on every canvas carries one of three states. The label is computed from where the content came from — a verbatim Ray, an entailed Source doc, or a Refract suggestion — so the model can never re-classify its own output as fact.

  • Said A verbatim Ray — the buyer’s own words, speaker-attributed.
  • Sourced Entailed by a Source document in your org knowledge base.
  • AI-suggested Refract proposed it. Rendered dashed everywhere it appears.

AI-suggested renders dashed everywhere — including customer surfaces. There is no path where a suggestion can disguise itself as something the buyer said.

Focus stage · customer view verified-only
Requirement Said

"Everything has to sync back to Salesforce on the Opportunity."

Capability Sourced

Artifacts + summaries push out to the Opportunity.

Proposed AI-suggested

Suggested: phased rollout, pilot team first.

Same surface the buyer sees — the dashed card stays dashed.
The verification gate

Customer-visible content passes an entailment gate first.

Before anything reaches the Focus stage, Refract checks that it is entailed by its source — the buyer’s words or a grounded doc. Content that isn’t entailed is held back, not shown. The SE curates on top of what passes; the gate never relaxes.

  • Entailed by a Ray or Source → reaches the stage
  • AI-suggested → may reach it, but stays visibly dashed
  • Not entailed → held in the workspace, never staged
candidates entailment gate Focus stage Said Sourced AI-suggested unentailed entailed? held in workspace — never staged verified-only
Auditable coverage · Wavelength

A number you can audit — not an LLM’s opinion.

Coverage is computed deterministically: SQL over a citations join table. Which discovery questions are answered, which facts made the solution — counted, not guessed.

questions cited 17 / 21
each segment links to the citing Ray
SELECT q.id,
       count(c.ray_id) > 0 AS answered
FROM   discovery_question q
LEFT JOIN citation c ON c.question_id = q.id
GROUP BY q.id;
Append-only fact ledger

Dismissals and supersedes are status — never deletion.

The ledger only grows. A fact can be dismissed or superseded, but it is never removed — so the record of what the buyer said, and what you did with it, stays auditable end to end.

  • ray_0142 active
    cited on Proposed Solution
  • ray_0151 superseded
    by ray_0188 · both retained
  • ray_0163 dismissed
    kept, marked out of scope
  • ray_0188 active
    cited on Architecture

rows only appended · the event log can replay any session

Shipped capabilities

Everything live on the dogfood tenant — no vaporware.

PrismaCue ships the full loop: capture to canvas to CRM to customer. Each capability below is in production. Items still in flight are on the roadmap — not listed here.

Capture

Capture & Ledger

  • Live meeting-bot capture via Recall and desktop SDK; accepts paste, dictation, and voice-note ingestion
  • Speech → Ray in ~15 s; end-of-call turn-merged transcripts with Ray re-anchoring
  • Beam Viewer: full transcript ↔ Ray-excerpts, every Ray anchored at its quoted span; double-click to jump to its moment in the call
  • Extraction disciplines validated on a real annotated call with a golden-corpus CI gate
Canvases

Solution Canvases

  • Template-driven generation with per-group budgets; new facts patch the canvas mid-call with no manual refresh
  • Unified expand-in-place cards with inline editing; categories selectable, movable, and resizable
  • Journey board — the cross-canvas overview for the full deal arc, with table-arranged categories carrying each card's full source label
  • Product-gaps canvas with strict conversation-grounding rules, capturing feature asks verbatim
People

People

  • Per-workspace person records auto-built from speakers as calls are captured — no manual entry
  • Person viewer accessible from any name tag anywhere in the product
  • Salesforce-backed person research with confirmable "Maybe" suggestions that you approve before they commit
  • Workspace inbox
Customer Surfaces

Customer Surfaces

  • Focus stage: presenter-driven, branded — only verified-and-curated content ever reaches the customer view
  • Recap composer with a first-class viewer for structured post-call summaries
  • Pretty PDF exports of canvases and recaps, each with a cited-Rays rail so every claim traces to the buyer's words
Salesforce

Salesforce

  • Connected-app OAuth — per-org setup, no shared credentials
  • Opportunity and account data seeds IN to seed the workspace with existing context
  • Artifacts and summaries push OUT to the Opportunity record after the call
Platform

Platform

  • Multi-tenant Postgres row-level security with CI probes — isolation verified continuously, not assumed
  • Ordered event log with replay: the cockpit can reconstruct any session from first principles
  • Per-tenant and per-deal token and cost telemetry on every model call
  • Degraded mode: LLM outage → capture-only; the ledger keeps running and nothing is lost

Version of record: product/TRUTH.md · 2026-06-12. Items in flight are on the roadmap and will appear here when shipped.

Built for your whole team

Every role gets the surface it actually needs

PrismaCue puts source-labeled facts in the hands of everyone who touches a deal — the SE, the AE, leadership, and product — each seeing exactly their view.

Primary persona
Sales Engineer

Ships the deliverable same-call. Never reconstructs from memory.

While you run discovery, PrismaCue assembles the solution canvas live — every card citing the buyer's exact words. By the time the call ends, the deliverable is already done.

  • Aperture — Question templates fill themselves as the customer answers — the cockpit asks ≤6 keys per call.
  • Ray capture — Every verbatim fact lands ~15 s after the words are spoken, speaker-attributed and kind-classified.
  • Spectrum canvas — Current State, Proposed Solution, and Architecture canvases assemble node-by-node, live.
  • Focus — Only verified content ever reaches the customer surface — the SE curates; the gate never relaxes.
Proposed Solution · assembling live live
Pain Said

"We lose two days rebuilding every discovery into a solution doc."

Capability Sourced

Proposed design maps to your published reference architecture — entailed by a Source doc in your knowledge base.

Proposed AI-suggested

Phased rollout, pilot team first — SE reviews before it reaches the stage.

Coverage
3 / 4 questions answered
Account Executive

Recap and mutual action plan grounded in the customer's own words.

The recap composer builds from the same source-labeled fact ledger. The pretty-PDF export carries every fact back to its quoted span. The Focus stage is presenter-driven — you control what the customer sees.

Sales Leadership

Deal coverage, gaps, and discovery quality — deterministic, not an LLM's opinion.

Wavelength computes coverage as SQL over a citations join table: which discovery questions are answered, which facts actually drove the solution. A number you can audit. Rollups across every deal in the team.

Product Teams

Feature asks captured verbatim from real calls, not paraphrased summaries.

The product-gaps canvas is governed by strict conversation-grounding rules — every card ties to the customer's exact words. No rounding, no abstraction between the customer's mouth and the backlog.

Go-to-market motion

The demo is the deliverable.

PrismaCue doesn't need a separate demo environment. It runs on its own calls, arrives at yours pre-seeded with the prospect's world, and ships the artifact live — every node traceable, every claim honest.

  1. Sell it with itself

    PrismaCue runs its own sales calls. The Navigator you're shown IS the product documenting that conversation — live Beam capture, source-labeled Rays landing ~15 seconds after the words are spoken, the Spectrum canvas assembling in real time. Not a recording. Not a mock. The actual product, running on its own pitch.

  2. Prospect Prebuild

    Before the first call, PrismaCue ingests the prospect's public corpus and seeds a Navigator for their business. Every node honestly badged AI-suggested, every claim one click from its source. You arrive knowing their world; they see the product already working on their problems.

  3. The 90-second artifact

    Rays land within 15 seconds of the words being spoken. The Spectrum canvas builds node-by-node, live. By the time discovery wraps, the deliverable — fully source-labeled, entailment-verified, ready for the customer Focus stage — is drafted. The demo ends and the document ships. Same call.

[Prospect Co] · Prebuild workspace all AI-suggested
Pain AI-suggested

Likely rebuilding solution docs after every discovery call — fidelity lost, momentum gone.

Context AI-suggested

SE team, quota-carrying, multi-product — inferred from public job postings.

Requirement AI-suggested

Salesforce integration — Opportunity field references appear across public materials.

Discovery Q AI-suggested

How does the team currently share solution artifacts after a call?

14 nodes seeded ready before call 1

See the 90-second artifact — live, on a real call

Book a call and watch PrismaCue run on that call itself. No separate demo env. No prepared data. The same product, the same source labels, the same live Spectrum canvas you'd use with your own prospects.

~15 s
spoken word → cited Ray
3-state
Said · Sourced · AI, in code
Verified-only
reaches the Focus stage
Pricing

Tailored to your team — let's talk.

Pricing is calibrated to team size, call volume, and the capabilities that matter most to you. The tiers below show what each plan emphasises — reach out and we'll find the right fit.

Team

Core SE productivity

Everything a sales engineer needs to ship the deliverable before the call ends — live capture to customer stage.

Pricing

Tailored — contact us

No public rate card yet

  • Live call capture — meeting bot + desktop SDK Beam
  • Ray extraction, speaker-attributed (~15 s on live calls) Ray
  • Aperture discovery workspace Aperture
  • Spectrum canvases — Current State · Proposed Solution · Architecture Spectrum
  • Focus customer stage — only verified content reaches the buyer Focus
  • Recap composer + PDF exports with cited-Rays rail
Most popular
Business

Team + CRM + deal analytics

Adds Salesforce, Wavelength deal-coverage analytics, and leadership visibility across every opportunity.

Pricing

Tailored — contact us

No public rate card yet

  • Everything in Team
  • Salesforce integration — Opportunity seeding in, artifacts + summaries out Beam
  • Wavelength coverage scoring (deterministic, auditable) Wavelength
  • Person records auto-built from call speakers
  • Workspace inbox
  • Leadership deal rollups + org dashboard soon
Enterprise

Platform + compliance + custom AI

SSO, BYO-AI model seam, per-tenant cost telemetry, and onboarding support for larger organisations.

Pricing

Tailored — contact us

No public rate card yet

  • Everything in Business
  • SSO — SAML / OIDC via WorkOS soon
  • BYO-AI model seam per org Refract
  • Per-tenant token + cost telemetry on every model call
  • Multi-tenant isolation (Postgres RLS with CI probes)
  • Custom onboarding + support

No prices on the page — intentionally.

PrismaCue is early and pricing isn't one-size-fits-all. Rather than post a number that doesn't fit your team, we'd rather have a 20-minute conversation, understand your call volume and workflow, and tell you exactly what it costs. Drop us a line and we'll get back to you quickly.

All capabilities above trace to shipped features as of 2026-06-12. Items marked soon are in active development.

The 90-second artifact

See PrismaCue build a canvas on a live call.

A solution canvas assembling itself in real time — every node honestly badged, every card citing the buyer's own words.

Request a demo

Spoken. Seen. Solved.