How Spark works
A morning ritual. A talkable filing cabinet. A network that stays warm.
Here's the whole product, end to end — how you set it up, how it shows up each day, and how it gets quietly smarter the more you use it.
Step 1 · Import
Bring your people in. Once.
Day one, Spark pulls in your existing network. From your phone's contacts (with the system picker — no full-disk access, no permission creep), or from a CSV your old tool exported (Follow Up Boss, kvCORE, HubSpot, anything).
CSV imports run through an AI mapper: Spark looks at your column headers and the first five rows, proposes a column-to-field mapping with confidence scores, and flags the ones you should double-check. You confirm, then watch the progress bar.
Duplicate handling is built in. If a contact comes in with a phone or email you already have on file, Spark merges instead of duplicating — and never overwrites your existing data, just fills in blanks.
Step 2 · Talk to Spark
Tell Spark. It files.
Every screen in Spark has a sparkle in the top right. Tap it and you're in a conversation with Spark. Type, or hold the mic and speak.
Tell it anything: "Jamie just had her second kid — a boy, Henry, born last Thursday." Spark writes a FAMILY note, adds an important date, and shows you what it did with one-tap Undo per action.
When you mention otherpeople — "Joe's brother lives in Denver," "Sarah and I both climb" — Spark files across the relationship graph. Both contacts get notes. So do you, when the fact is about you.
If multiple people in your network share a name, Spark doesn't guess. It surfaces a tap-to-pick chooser inline ("which Sarah?") and resumes once you've answered.
Step 3 · The morning list
Three to seven things worth doing today.
Every morning at 7am local, Spark runs a job: it looks across your AI-tracked contacts, scans birthdays in the next 30 days, checks due reminders, evaluates cadence windows, picks topics, and generates the curated short list for today.
You get one push: "Today is ready." Open the app. Three to seven things, ranked by how much they actually matter today. Each one has a drafted message ready to go.
No inbox. No backlog. Tomorrow's list will be fresh, and yesterday's unacted-on suggestions quietly go stale at midnight. Inaction is data, not debt.
Step 4 · Voice
Spark writes the way you actually do.
Most AI-drafted messages read like a LinkedIn comment. Spark's don't. Once a week, Spark studies the messages you've actually sent — through Spark, not from your full email — and distills a small profile of how you write.
Tone (warm, direct, playful, formal). Length (brief, medium, long). Sign-off (cheers / a single first initial / nothing at all). Emoji habit. A handful of phrases you actually use.
Every draft Spark writes runs through that profile. The output sounds like you talking — not like an assistant pretending to be you.
Step 5 · News-grounded · Pro
"Heard about the Ravens trade?"
When Spark drafts a check-in to a contact who's a Baltimore Ravens fan, it knows the Ravens just traded for a new wide receiver. When it drafts to someone who works at Stripe, it knows they shipped a new product this week.
Here's how. Spark extracts topicsfrom each contact's notes — the Ravens, Stripe, the Catholic Church, marathon running, whatever you've told it. Once a day, a search runs across the union of every Pro user's topics (one search per topic, deduped — so 200 Spark users with a Ravens fan share one Ravens search).
At draft time, Spark quietly checks if there's fresh news on any of that contact's topics. If there is, it weaves it in. If not, it drafts the normal way.
Sentiment-filtered. Spark never drafts "thinking of you" because a tragedy hit their hometown — it hands you that signal differently, and only if you want it.
Step 6 · Learns from no
The smarter half of the loop.
Dismiss a suggestion and Spark asks why. Pick a tag — tone wrong, timing wrong, wrong channel, already in touch, not interested. Or type the reason out.
Spark turns that into a durable fact about the contact. "Steve prefers email." "Anna's a Steelers fan, not a Ravens."Those facts show up as "What Spark learned" chips on the contact, and they shape future drafts and suggestions.
After three consecutive rejections for the same contact in one generation cycle, Spark backs off and doesn't surface them again until the next cadence window opens or something genuinely new happens.
Most apps treat "no" as silence. Spark treats it as the most valuable signal you give it.
Step 7 · Afternoon questions · Pro
Spark fills in its own gaps.
Every weekday at 3pm local, Spark looks at the contacts whose file is thinner than it should be — the ones you marked as close, but Spark doesn't actually know much about them yet — and writes one specific question per person.
"What's John's daughter studying again?" "Where does Priya live these days?" Tap the row and the Steward opens with that question as Spark's first line. Type or speak the answer — Spark files it. Bio refreshes. Future suggestions get smarter.
Up to three questions a day. Per-contact cooldown of two weeks. Spark won't ask if it doesn't have something specific to ask about.
Step 8 · Trust
Every action is yours to undo. Every decision is yours to see.
Every AI action Spark takes is reversible for 24 hours. Wrote the wrong note? One tap. Filed against the wrong contact? One tap. Generated a draft you hate? Regenerate, edit, or just don't send.
Every AI call is logged. Settings → Spark activity shows a chronological list of everything autonomous Spark has done — bios refreshed, notes filed across contacts, suggestions generated, topics extracted. Tap a row to jump to the contact it affected.
Every Steward conversation persists in full — including the tool calls Spark made, what they wrote, what their inputs and outputs were. Nothing about what Spark does is opaque.
Ready to try it?
14-day Pro trial. Cancel anytime. No credit card needed to start the free tier.
Coming to the App Store soon.