Bryn watches. You decide the Play. Bryn runs it.
One loop, on the high ground: watch to score to run to learn. You define the Play and approve it once. Bryn runs it every time it sees the pattern, without waiting on you.
Nothing signals alone.
Three surfaces, one scope: your product, your site, your CRM. Bryn watches them together, including the anonymous visitors most tools miss. The last sixty minutes on one timeline, every signal, the hits scored as they land.
The score shows its work.
It scores the intent against your ICP, not against a generic "high intent" label. Fit is ICP shape: does this account look like the ones that close. Intent is signal density and velocity in the window. Timing is recency. Each axis is a click-through to the contributing signals; the composite is the answer. Weights are yours to tune from Growth upward.
The Play fires, end to end.
It runs the Play you approved, into your stack (Slack, your CRM, your outbound), within the same minute. One signal can route to several destinations; different Plays route differently. Bryn executes; you authored what to execute.
Bryn learns by recording.
Every run is recorded in the audit log, the work record your team and your CFO can both read. Your data stays yours. Bryn records what it did, not who you are. Bryn sharpens from the real decisions you make in the flow of work: your edits to scoring weights and Play definitions, and the outputs you dismiss. Today that sharpening is your hand on the weights, and the recorded log is what makes automatic re-weighting possible later.
Bounded on purpose.
Bryn's autonomy is bounded by approved Plays. The restraints are the feature.
No action outside Play-defined channels.
Bryn writes only to the systems your Play names, and never makes unbounded, open-ended API calls.
At the cap, Bryn pauses.
New identifications pause; existing Plays keep running on accounts already in scope. An in-product prompt offers the upgrade. No surprise bills.
One click, and Bryn waits.
Suspend all execution, or cut by region, segment, or Play. The audit log captures the suspend. Bryn waits. You decide when.
You hold the Play. Both ends of it.
A Play is an operator-defined, operator-approved action pattern: when this kind of signal appears on this kind of account, run this sequence into these destinations, with these guardrails. Authority lives at Play definition and Play approval, not at per-instance execution. Run mode is the default; Approve mode is opt-in for sensitive Plays; the kill switch is always one click.
Bryn watches. You decide the Play. Bryn runs it.
Run mode. Approved Plays execute end to end, within the minute, on the record.
Bryn answers questions where you already work.
Bryn's record and its scope are exposed as tools in any MCP client. From Claude desktop or Cursor, ask about a pattern:
pricing → comparison → repeat (7d)
Bryn answers from the same audit log the product shows you: which accounts matched it this week, the Play that fired on each, the composite score behind every decision, and the events that moved it.
Connect Bryn →9 accounts matched the pattern this week. 6 routed, 3 held below threshold. The demo-intent Play fired on the 6 in Run mode.
Top composite was 81 on acct_8112 (Fit 84, Intent 79, Timing 71), routed to #growth-signals, a CRM task, and the comparison landing page. Each step is in the audit log, filed 14:02 UTC, source-traced and replayable.
Every signal, score, decision, and run: on the record.
Time-stamped, source-traced, replayable. The log is your work proof and the substrate for what Bryn learns next. Your CFO and DPO inherit the same trail. Retention windows, export, and erase-on-demand live in the trust center.