Vyntric Cadence user guide
Cadence is an AI product owner for a fractional executive who runs product across several client companies at once. It ingests business context from the places work actually happens, turns that context into a live understanding of each client's product, and does the product owner's job in Linear while keeping you situationally ahead of every relationship.
This guide is the single source of truth for product documentation. It grows in the same commit as any user-facing change.
Getting started
- Sign in at the root page with your allowlisted email.
- Connect providers under Connections: Slack (session tokens), Google (OAuth), Microsoft (OAuth), Linear (personal API key per workspace), and Granola.
- Create a client workspace per client company and map sources to it.
- Write a context note for each client: active projects, worries, what a win looks like this quarter.
- Review the playbook and autonomy settings; the standard defaults are sensible and every value is overridable per client.
- Add voice samples (three short-form, two long-form) so drafts sound like you.
- Set your briefing destination (default: a Slack self-DM in an owned workspace).
The daily loop
Every day at your configured sweep hour, Cadence reads a rolling window from every connected source, resolves items to clients, extracts typed product signals, generates proposals under each client's autonomy settings, and writes your briefing. You read the briefing, decide proposals, and review suggested outreach. Cadence never sends a client-facing message on its own.
Clients
Each client workspace has its own connections, sources, playbook, autonomy
settings, inbound email address ({client-slug}@withcadence.so), and
history. A failure or pause in one client never affects another.
Proposals and autonomy
Every Linear-writing behavior belongs to a category (annotate, create_issue, triage, structure, plan, status) with a per-client autonomy level: observe, propose, or auto. Propose is the default for everything except context comments and status updates. Rejecting a proposal with a reason turns that reason into a steering note that shapes future behavior.
Cycle planning
On your planning day, Cadence composes the next cycle from issues carrying the ready label, against capacity measured from your trailing velocity, and shows every exclusion with its reason. Delivery projections always state p50 and p85 with the assumptions attached, and refuse to project at all when there is no velocity history to be honest about.
The MCP endpoint
Mint a personal access token under Tokens, then add
https://withcadence.so/mcp to Claude Desktop, Claude Code, or any MCP
client. The same operations are available over REST at /api/v1 with the
same token.
Reading the briefing well
Every claim in a briefing cites an observation from that window. Where a source was unreadable, the section carries a coverage note instead of calling the client quiet. A section that says nothing is a real answer, not a failure.
Proposals
The proposal queue is the main place you work. Each proposal cites the signals that motivated it, shows its prioritization math, and, for issues, the definition-of-ready check. Approving executes it against Linear under your own name. Rejecting with a reason turns that reason into standing guidance that shapes future behavior.
Shaping a feature
Hand off a raw idea and walk away. Cadence drafts an epic, milestones, and issues against your conventions, decides everything it can ground in context, and asks the client's team about what it genuinely cannot decide, in a Slack channel you configure and acknowledge. When answers conflict it names the conflict rather than picking a side. Issues whose questions stay unanswered are created without the ready label, carrying those questions in the body, so an incomplete definition is visible rather than silently blocked.
Directives
Delegate long-running product work in natural language: "clean up this Linear workspace, organize issues into projects by feature, prioritize, and get everything ready to work." Cadence plans it, runs it in the background, and reports back over Slack DM when the changeset is ready for review.