How Klyr works, end to end.
Klyr turns customer evidence into ranked product bets, builder-ready handoffs, and a product memory that compounds. This guide covers the workflow, the concepts, and the questions teams ask most.
From a blank workspace to a shipped, reviewed bet.
The fastest way to understand Klyr is to run one full loop. The sample project lets you do it without uploading anything of your own.
Create an account
Open the workspace and sign up with email and password. You start on the free Starter plan with a workspace of your own. No card required.
Start from the sample or your own evidence
Load the sample project to see a finished evidence-to-build loop, or go to Research and paste a transcript. Klyr accepts pasted text and .txt, .vtt, .srt, and .docx files.
Run a synthesis
Pick Balanced (the default) or Highest quality mode and run. Klyr extracts themes ranked by impact, each backed by verified citations into the exact transcript spans that support it.
Ask what to build
Use the PM workbench to ask the question your team is debating. Klyr returns ranked opportunity plans grounded in your evidence. Save the strongest plan to a project to create Feature Bets.
Generate a Build Pack and hand off
Turn a Feature Bet into a Build Pack: PRD summary, acceptance criteria, implementation tasks, evidence citations, and a coding-agent prompt. The Spec Quality Gate and Decision Room keep weak specs in draft.
Record the outcome
After the work ships, mark the bet shipped and capture expected-versus-actual learning. The result becomes Product Memory, so the next recommendation knows what actually worked.
Eight objects, one chain of evidence.
Every object in Klyr exists to keep one chain intact: what customers said, what the team decided, what got built, and what happened next.
Evidence
The raw material: interview transcripts, sales notes, support themes, customer feedback, imported product data, and uploaded recordings with transcripts. Everything else in Klyr links back to evidence.
Themes and citations
Synthesis output. Each theme states a problem, its frequency, severity, and confidence, and cites the exact quotes that support it, with timestamps when the transcript has them.
Feature Bets
Durable, ranked product bets created from saved opportunity plans. A bet captures the problem, audience, evidence, counter-evidence, confidence, expected outcome, risks, and implementation outline.
Build Packs
One-click handoff bundles tied to a Feature Bet: PRD-style summary, acceptance criteria, task plan, UI/data/API changes, evidence citations, non-goals, risks, and a coding-agent prompt.
Spec Quality Gate
Before handoff, Klyr scores each bet for evidence depth, counter-evidence, outcome metric, implementation clarity, and prompt quality. Weak specs stay draft; ready specs go to builders.
Decision Room
A lightweight review layer for cross-functional sign-off. Founder, PM, design, engineering, and sales reviews stay simple: pending, approved, changes requested, or not needed.
Product Memory
The compounding record of shipped bets, misses, proven wins, and prior tradeoffs. The Evidence Graph makes the chain inspectable: quote → theme → Feature Bet → Build Pack → outcome.
Outcome learning
The closing step of the loop. Shipped bets get reviewed against their expected outcome, and what the team learned is saved where the next decision can find it.
One loop from raw evidence to shipped outcome.
Upload evidence
Bring in interviews, feedback, sales notes, support themes, and product usage context. Paste text, upload transcript files, attach recordings with transcripts, or import product data as CSV. All from Research or a project’s Evidence section.
Ask what to build
Run a synthesis, then ask the question the team is really debating. Klyr answers with ranked product bets, each grounded in citations, confidence, risks, and counter-evidence, never opinion alone.
Generate the plan
Save the strongest plan to a project. Feature Bets carry the decision; cited PRDs, decision briefs, and contradiction checks give the team something to inspect instead of debate.
Hand off to builders
Generate a Build Pack, pass the quality gate and Decision Room review, create the implementation tasks, then copy the builder brief or the Klyr context pack for the people (or coding agents) doing the work.
Learn from outcomes
Mark the bet shipped, record what changed and whether the expected outcome happened, and let Product Memory carry that learning into the next product bet.
Retrieval first, then generation. Always.
Synthesis runs a five-stage pipeline designed so that every bullet in the output ties to a source span you can inspect.
And nothing in your workspace mutates on AI's say-so: agent runs, automations, and Command Center actions propose changes that you review and accept first.
Sample and propose
The model scans a representative slice of your transcripts and proposes candidate themes, each with a search query.
Retrieve
Every transcript chunk is embedded, and the strongest candidate quotes for each theme are found by similarity search.
Verify
A separate model pass checks each retrieved quote and rejects any that do not actually support the theme. No verified quote, no claim.
Gate
Themes with fewer than two verified citations are dropped, and when you upload multiple transcripts, support must come from at least two of them. Better four strong themes than twelve weak ones.
Spec
Surviving themes get user stories and acceptance criteria, so the output is ready for planning, not just reading.
Confidence labels follow the citations: a theme with six or more verified quotes reads high, three to five reads medium, and anything thinner reads low, so you always know how much weight a theme can carry.
See a live cited reportWhere everything lives.
Projects
The portfolio view and each project workspace. This is where Feature Bets, Build Packs, tasks, docs, and outcomes live.
Research
Run syntheses, ask research questions, generate PRDs and tickets, find contradictions, and score opportunities against your evidence.
Library
Saved research runs and evidence, ready to link into projects and revisit later.
Inbox
Team notifications: mentions, task assignments, blocked work, and due or overdue dates.
Workspace
Team settings, members and roles, plan and billing, and AI usage summaries.
Import
The Migration Center. Bring tasks in from other tools with editable field mapping, and import product data CSVs that can support or refute your bets.
Help
Send a bug report, question, billing note, or data request with workspace context already included.
⌘K command palette
Press ⌘K (or Ctrl+K) anywhere in the dashboard to search the workspace, jump to any section, run review-first AI presets, create tasks, link evidence, generate Build Packs, and start outcome reviews.
Overview
The Project OS guide: a state-aware view that always points at the next useful action. Add evidence, ask what to build, generate a Build Pack, review shipped learning, or unblock work. Includes the team setup checklist and activity timeline.
Work
Tasks in board, list, and timeline views with owners, status, priority, dates, effort, subtasks, checklists, and threaded comments with @mentions. Sprint cycles live here too.
Evidence
Everything the project’s decisions rest on: linked research, transcript syntheses, uploaded files and recordings, and imported product signals.
Docs
Editable workspace documents. PRDs, decisions, launch briefs, weekly updates, meeting notes, and wiki pages, with templates and revision history.
Outcomes
Goals, outcome reviews, and the learning captured from shipped Feature Bets.
Share
Generated project reports and share links for stakeholders who live outside the workspace.
Two modes, no surprise bills.
Every plan includes monthly AI runs split between the two modes. Usage stays visible in Workspace settings, and limits per plan are on the pricing page.
See plan limitsBalanced
The everyday path for syntheses, research questions, PRD drafts, tickets, and contradiction checks. Cost-controlled so you can ask more questions without rationing.
Highest quality
The premium path for deeper evidence passes, high-stakes reports, and decisions where the team needs the strongest reasoning. Intentionally limited per month on each plan.
Evidence in, builder-ready work out.
Klyr meets your existing stack with imports on the way in and copy-ready, evidence-backed payloads on the way out.
Transcripts
Paste text directly, or upload .txt, .vtt, .srt, and .docx files. Speaker labels and timestamps are kept when present and show up in citations.
Recordings and large files
Upload specs, designs, images, PDFs, spreadsheets, and recordings to projects and tasks. Large media uploads are resumable, files open through secure signed links, and you can paste a transcript onto a recording to use it as synthesis evidence.
Product data (CSV)
Import analytics, activation funnels, churn reasons, support tickets, and sales or customer-success notes. Matched signals appear directly on related Feature Bets as support, risk, or context.
Task imports
Move work in from other tools through the Migration Center. Imports are inspectable row by row, field mapping is editable before confirmation, and runs are reversible while you test a cutover.
Tool handoffs
Copy evidence-backed payloads formatted for GitHub, Linear, Jira, Notion, Slack, and email. Native GitHub and Linear issue creation is available when server-side credentials are configured.
Agent context
Every project exposes a read-only, machine-readable context endpoint, and Build Packs include a copyable Klyr context pack, so coding agents can consume the same evidence-backed state your team sees.
The questions teams ask first.
What can I upload as evidence?
Pasted text and .txt, .vtt, .srt, and .docx transcript files for synthesis; CSVs for product data; and project file uploads (documents, designs, images, spreadsheets, recordings) that can carry transcripts into synthesis. Timestamps and speaker labels are preserved when your transcript has them.
Why did a theme I expected not show up?
Every candidate quote is verified by a separate model pass, and themes with fewer than two verified citations are dropped, with support required across at least two transcripts when you upload more than one. If a theme is missing, the evidence for it was too thin. Add more source material and run again.
What counts as an AI run, and what are the limits?
Synthesis runs and PM workbench requests count against your plan’s monthly limits, split between Balanced and Highest quality modes. Smaller artifacts like PRD drafts and contradiction checks are far cheaper than full syntheses. Current usage is always visible in Workspace settings, and plan limits are on the pricing page.
Will Klyr change my tasks or documents on its own?
No. Klyr is review-first everywhere: Command Center proposals, Klyr Agent Mode runs, automations, and PM health scans all produce proposed operations that you inspect and accept before anything in the workspace changes.
Can I export or share what Klyr produces?
Yes. Synthesis reports export as Markdown, projects generate shareable reports and share links, and Build Packs include copy-ready builder briefs, tool payloads, and coding-agent prompts.
How do teams work in Klyr?
The Team plan adds seats, workspace invites, and admin, member, and viewer roles. Teammates share projects, evidence, Product Memory, and the Inbox, and task comments support @mentions that notify the right person.
Where does my data live, and how do I delete it?
Workspace data is stored in a Postgres database and private file storage, scoped to your workspace with row-level security. You can request an export or deletion any time through the data request page or from Help inside the dashboard.
How do I get help?
Open Help in the dashboard sidebar to send a support request with workspace context attached, or email Support@GetKlyr.io.
Something missing? Open Help inside the dashboard or write to Support@GetKlyr.io.
The fastest tour is one real synthesis.
Open the workspace, load the sample project, and run the loop once. Everything in this guide will make sense in about five minutes.