The Blueprint

One brain that finds, writes, sends, and learns — for you.

ASTRA is a revenue operating system for founders, freelancers and small teams. It replaces the messy stack of AI generators, CRMs, sequencers, and coaches with one loop that compounds every time you use it.

The problem

AI tools are dumb generators
Generic outputs, no memory, no improvement from your wins.
CRMs are storage, not selling
They demand data entry and give back nothing useful.
Sequencers blast, they don't sell
One-size-fits-all cadences that burn domains and lists.
Agencies cost $5–25k/month
Out of reach for the operators who need selling most.

The result: brilliant operators stay invisible. Their voice never lands consistently, follow-ups slip, and they lose deals they could have won. The playing field shouldn't be reserved for people who can afford a sales team.

The moat

Compounding memory
Every interaction improves the next one — per-user, not per-session.
One voice across surfaces
Sales, marketing, follow-ups — same Brand Kit, same context.
Opinionated defaults
We ship a point of view. Not another blank prompt box.
The Review Ritual loop

One rhythm. Four moves. Compounds every week.

Pick the day you review pipeline — with your manager, or with yourself. ASTRA anchors the whole product to your review day, in your timezone.

  1. 1Commit

    Tap Commit on any lead.

  2. 2Queue

    Sorted by your review day.

  3. 3Nudge

    Day before: what's still open.

  4. 4Recap

    A scoreboard worth screenshotting.

User journey — Day 0 to Day 90

Day 0
Onboarding, in minutes
Connect mailbox, paste your website, upload one past win. ASTRA reads your voice, builds the Brand Kit, and pulls in your recent contacts.
Day 1
First win
Paste a target domain. ASTRA returns a shortlist of the right humans and a first-draft email that sounds like you. You edit a line and send.
Day 3
First reply, first follow-up
One prospect replies, others go quiet. ASTRA drafts follow-ups for the silent ones — different angle each. You approve in a tap.
Day 7
The loop kicks in
A morning brief lists signals matched to your list, replies overnight, and meetings scheduled — so your first move each day is obvious.
Day 30
Autopilot on a segment
You trust it on a low-stakes segment (past demo no-shows). ASTRA runs the sequence; you review a weekly digest.
Day 90
Team upgrade
Bring in a teammate. They inherit the memory and are useful on day one because ASTRA already knows the accounts.

Every feature, in plain English

What it is · what it solves · how it works · what powers it.

Find contacts

What: Paste a domain (stripe.com, zomato.com) — ASTRA returns a shortlist of the right humans, not a scraped list.

Solves: Cold prospecting eats hours a week and returns a spreadsheet no one reads.

How: Domain → firmographic + role inference → ranked shortlist with why-this-person reasoning.

Tech: Web enrichment + LLM ranking with dedupe against your CRM history.

Enrichment & memory

What: Every contact carries context: role, past threads, what they replied to.

Solves: You forget who said what. Generic follow-ups get ignored.

How: Persistent per-contact memory that survives sessions and compounds with each interaction.

Tech: Supabase + pgvector for semantic recall; RLS-scoped per user.

Draft & personalize

What: First draft in your voice, tuned to the recipient — ready in seconds, not hours.

Solves: Blank page. Generic AI output that reads like AI.

How: Your brand voice + memory + signal + template → draft you'd actually send.

Tech: Voice profile stored in Brand Kit; prompt composition per-send.

Schedule & auto-send

What: Queue sends for the right hour in the recipient's timezone. Auto-throttled to protect deliverability.

Solves: Manual send times, batch blasts that torch your domain reputation.

How: Per-recipient timezone inference + warmed sending windows.

Tech: pgmq queues + cron worker; per-mailbox rate limits.

Follow-ups on autopilot

What: If they don't reply, ASTRA drafts the next touch — different angle, not the same nag.

Solves: Most deals need several touches. Most people stop at one or two.

How: Reply detection → variant selection → scheduled next step until a stop condition.

Tech: IMAP reply parsing; sequence state machine per lead.

Reply Intelligence — opt-in

What: Off by default. When you opt in, ASTRA classifies every inbound reply (interested / pricing / timing / objection / OOO) and surfaces reply patterns so drafts stop being generic — you sleep, ASTRA works, with your consent.

Solves: Rival tools silently read and store every reply body. That's the wrong default. Consent is the moat.

How: Explicit opt-in in Settings → Privacy & AI. Store only a 280-char preview + subject. Auto-purge at 30 days. One-click revoke + purge in the same tab. Every consent event audit-logged.

Tech: user_ai_preferences + ai_consent_events tables; Gemini classification with no training and no cross-user use; nightly purge cron.

Brand Kit

What: Your voice, tone, offer language, colors, logos — set once, read into every draft.

Solves: AI output that sounds like everyone else's AI output.

How: One-time setup during onboarding; every draft reads from it.

Tech: Per-workspace brand profile; version-controlled.

Marketing Studio

What: Long-form posts, short-form hooks, newsletters, sales pages — same voice, same memory as your outbound.

Solves: Marketing and sales say different things in different voices.

How: Studio pulls from the same Brand Kit + memory as Sales Buddy.

Tech: Shared context layer across surfaces.

Autopilot

What: Turn on Autopilot for a segment — ASTRA finds, drafts, schedules, follows up without asking.

Solves: You want revenue without becoming a full-time SDR.

How: Guardrails (daily send cap, forbidden domains, human-approval threshold) + run.

Tech: Policy engine + queue worker; every action logged for review.

Shared team memory

What: In Sales Buddy and Marketing Studio, teammates share one memory: who owns which account, who replied, what to never repeat.

Solves: Two reps email the same contact with contradictory offers.

How: Workspace-scoped memory with per-lead ownership.

Tech: RLS by workspace_id + role checks via user_roles.

Deliverability

What: SPF/DKIM/DMARC setup guide, warm-up guidance, per-send hygiene rules.

Solves: Great copy dies in spam folders.

How: Setup guide in Settings + per-send hygiene rules. No live per-domain audit today.

Tech: Static guide + rate limits + suppression list.

Analytics that mean something

What: Reply rate — not vanity opens. Meeting rate and revenue attribution on the roadmap.

Solves: Dashboards that make you feel busy without telling you what to change.

How: Every send tagged; every reply threaded; conversions written back.

Tech: Event log in Postgres; per-workspace dashboards.

Swipe to explore →

How it's built

Frontend: TanStack Start on React 19, Tailwind v4, SSR on Cloudflare Workers.

Backend: Supabase (Postgres + RLS), pgvector for memory, pgmq for send queues, cron workers for signals and follow-ups.

LLM layer: Model-agnostic — routing per task (fast for classification, quality for drafting) via a gateway.

Security: Row-level security on every table, role-based access via a dedicated user_roles table, no roles on profiles.

What we can't do yet

We publish this. Honestly. See the current gaps →

See it end-to-end

The demo walks you from cold start to a compounding loop in 5 acts. Or book 20 minutes with the team.