Built for Australian brands · Phase 0

The brand brain that remembers what worked.

HypaBrand connects your channels, learns from every result and turns evidence into approved action.

Open portal See how it works
Home: the brand overview with the KPI strip, what changed and the review queueHome: the brand overview with the KPI strip, what changed and the review queue
Sample data

How it works

One loop. One memory.

Every stage reads from the Brand Brain and writes evidence back into it.

Official APIs only. Read scopes first; tokens go straight into the vault.

Sample data

Improve feeds the Brain. The next Connect starts from a smarter brand.

Brand Brain

Eight kinds of record, never one bucket.

Grade, confidence and evidence travel with every record into every screen and prompt.

  1. Fact

    Something sourced or asserted, with its as-of date. A person verifies it.

  2. Observation

    A measured change against a baseline, detected by deterministic code.

  3. Hypothesis Proposed by Strategist · awaiting a person

    A face in the first frame stops the scroll

    Untested. The test that would check it: a randomised posting calendar.

    G0 · Unsupported Confidence · Low
    Predicted direction
    Increase
    Minimum effect
    20 %
    Primary metric
    Views at +24 h
    • Mechanism A face in the first frame
    • Proposed test Randomised posting calendar

    A candidate cause, stated so it can be tested. Never promoted by repetition.

  4. Decision

    A choice a named person made, with the evidence cited.

  5. Experiment

    A pre-registered test with a stated estimand. A person launches it.

  6. Result

    The computed outcome, with its interval. Only the evaluator writes one.

  7. Learning

    What the brand now knows, graded and linked to its results.

  8. Rule

    Machine-readable policy that Brand Guard, the ledger and the Analyst enforce.

Evidence grades, by rule

  1. G0 Unsupported. An idea, a suggestion or a platform recommendation, untested.
  2. G1 Anecdotal. One post, one comment, one campaign. Illustrates, never justifies.
  3. G2 Descriptive. Measured aggregates, driver decompositions, uncontrolled before/after.
  4. G3 Associational. Comparisons meeting minimum samples with the confounder checklist.
  5. G4 Quasi-experimental. Designs with passed diagnostics; calibrated models.
  6. G5 Randomised. A pre-registered randomised test with a stated estimand.
  7. G6 Replicated. Two or more independent G5 results in the same direction.

Who holds authority

People verify facts, make decisions, launch experiments, accept learnings and activate rules. Agents only propose.

Confidence moves a claim down, never up. Nothing earns causal wording except the design of the test.

Analyst

Answers why, without inventing numbers.

Deterministic analyses run first; the model only narrates what they found.

  • Runs pre-specified, deterministic analyses instead of improvising.
  • Grades its own claims by rule; wording follows the grade.
  • Says “insufficient evidence” plainly, with what it would take to decide.

What changed this week?

Reach rose 18 % Instagram on Instagram. Store revenue was flat at $12.4k STORE . Meta reports 42 conversions META against 27 in GA4 GA4 . The gap sits inside its usual band, so no action is proposed.

Numbers come from tools, never from the model. Four tool outputs · graded G2 Descriptive · sample data

Measurement

Five lenses, kept apart.

Every source tells a different story about the same money. You see all of them.

  • Meta-attributed Credits itself with conversions it may not have caused.
  • GA4-attributed Under-counts after consent loss and ad blockers.
  • Store Knows the revenue but not the cause.
  • Survey Customers misremember how they found you.
  • Blended Hides which channel works; only where the registry allows.
Conversions from one campaign, by lens Sample data
0102030405042Meta27GA431Store19Survey34Blended0102030405042Meta27GA431Store19Survey34Blended

Lenses are never merged.

Agents

Agents propose. People approve.

No agent calls a provider write API. Approved actions run through deterministic code.

  • Every agent has a written charter, a fixed toolset and an evaluation suite.
  • No autonomous budget, bid or status change until you enable a policy for it.
  • Approved actions carry a policy check, an audit row and rollback data.
Budget change Low risk Strategist · Expires in 2 days

Move $40 a day from Prospecting B to A

A beat B on store revenue per dollar for 14 days, above minimum sample. Store

Field Before After
Prospecting A · daily budget $120 $160
Prospecting B · daily budget $120 $80
Status Active Active

Actions: Approve, Edit, Reject (illustration).

Approved actions run through the executor with a policy check, an audit row and rollback data.

Security

Assume a breach. Limit its reach.

HypaBrand can spend money and speak as a brand, so no component can reach everything.

  • Forced row-level security

    Every tenant row carries its workspace; policies fail closed without a context.

  • Worker holds the keys

    The web tier never holds a key and can never decrypt a token.

  • Raw first, with provenance

    Every provider payload is kept before normalisation, so figures recompute from source.

  • People approve every action

    Agents propose; a deterministic executor acts with policy checks and rollback data.

Architecture: browser, web, worker, vault, Postgres and providers Web app · no secrets Worker · no inbound connection Browser Web Hono · sessions · CSP Worker jobs · connectors · gateway Postgres · RLS forced Tenant context, per transaction Vault Key-encryption key Providers OAuth code, sealed to the worker Official APIs, least privilege
Two applications, one secret boundary. The orange path is the sealed hand-off.

Read the security model

Integrations

Official APIs, read first.

Write scopes only when a feature needs them. Every data-use policy is enforced by a registry.

  • Instagram

    Organic

    Profile, media and insights through the Graph API.

  • Meta Ads

    Paid

    Structure, insights and diagnostics; changes prepared paused.

  • Google Analytics 4

    Google Analytics 4 Site

    Standard report cubes; model use only after consent.

  • Shopify Shopify

    Shopify Commerce

    Orders, products and analytics from a custom app.

  • LinkedIn

    LinkedIn Reporting only

    Post analytics shown live; values cached 48 hours, references only.

Ready when your brand is.

The portal opens without a sign-in while the pilot runs.

Open portal