Olimpo
BAST · Olimpo · v1

We turn ideas into impact.

From insight to outcome, in spectrum.

The Olimpo is a multi-agent creative system. One loop, two gates, six ofertable packages — inspired by SDD, adapted to creative production, progressively formalized run by run.

Explore the system → See how it works
v1 · 2026-05-04 6 packages 1 paradigmatic case Inspired by SDD

Multi-agent intelligence

Specialized agents collaborate to research, plan, build, and verify before output.

Validated by design

Every artifact is checked against contract before it leaves the loop.

Built for scale

From one-off mockups to always-on programs — same cycle, same vocabulary.

Connected to your stack

Drop-in deployable. Cloudflare, n8n, Supabase, WhatsApp — channels already open.

Layer 01 · The Cycle

One loop. Two gates. Six artifacts.

Every run starts as a User Story and ends as a Published Feature. Two gates — Verifier and Coherence Check — keep the work from drifting. The first asks does this match the spec? The second, only emitted by Ariadna, asks does the whole sustain the contract?

The Olimpo cycle: from User Story to Feature Published. 01 · INPUT User Story 02 · DECISION Decision-closed Refinement 03 · GATE — ARIADNA Contract 04 · SPEC Implementation Spec 05 · ITERATION LOOP Implementer → Verifier → fail/pass Lime · Nyx · Sibila Brigid Cross-model · QA visual 06 · ARTIFACTS BRANCH TESTS DOCUMENTATION CODE TESTING REPORT PROPOSAL UPDATE GATE — ARIADNA Coherence Check Feature Ready Feature for PR Feature Published

The spectrum system

Six packages running on one cycle. Each color is a function, not decoration. All converge back into white.

Strategic Sprint Mockup Web + Live Deploy Live Design System WhatsApp Agent Automated Survey Ad Creatives

Inside the system

Six specialized agents, indexed memory, canonical SOPs. The loop runs because the layers underneath already do.

01 · INPUT
User Story
The client's ask, translated to one sentence: as [ICP] I want [output] for [outcome].
02 · DECISION
Decision-closed Refinement
Explicit decisions cleared before a deliverable is promised. What's open gets flagged, never assumed.
03 · GATE — ARIADNA
Contract
Narrative coherence + visual cohesion + strategic alignment. The semantic north of the package.
04 · SPEC
Implementation Spec
The contract concretized in tokens, components, copy, motion, schemas. What the implementers execute.
05 · LOOP
Implementer → Verifier
An agent produces. A different agent verifies against spec and blocks if it fails.
06 · ARTIFACTS
Six auditable outputs
Branch · Tests · Documentation · Code · Testing Report · Proposal Update.
07 · GATE — ARIADNA
Coherence Check
Only Ariadna emits this. Blocks even if every artifact passed verifier — when the whole stops sustaining the contract.
08–10 · OUTPUT
Feature → PR → Published
Ready (local passes) → For PR (public preview) → Published (client received).
Layer 02 · The Packages

Six offers running on the same cycle.

Each package is the same cycle with a different contract. Same vocabulary, same gates, different ICP and output. The number isn't fixed — these are the six already validated against real runs.

01 Strategic Sprint + Founders / hiring managers who posted a senior role and want to see real capacity before inviting to interview.

Output

Strategic diagnostic + 1 live landing (public URL) + cold pitch / application.

Ariadna's Contract

The proposal sustains a "design partnership long-run" frame. Diagnosis is auditable claim, not flattery.

Verifier loop

  • Lime (visual) → Brigid (cross-browser + brand defense)
  • Sibila (copy) → Ariadna (frame check)
  • Nyx (build) → Brigid (deploy + accents + reduced-motion)

Acceptance criteria

  • Public URL sustains cold-email claim
  • Diagnosis doesn't contradict target's positioning
  • Zero invented data

Validated in

POLY · ESIM (paradigmatic) · Bitpanda · Convergent · Mercor (in flight) — Skill /sprint · SOP SOP_STRATEGIC_SPRINT

02 Mockup Web + Live Deploy + Solo-founders / pre-seed startups that need to show something viable to investors or users now, without fighting frameworks.

Output

Live site at <slug>.pages.dev with preview per iteration + reusable base tokens.

Ariadna's Contract

The page communicates the product promise in 8 seconds without off-page explanation. Mobile-first real, not responsive-as-joke.

Verifier loop

  • Lime (wireframe) → Ariadna (mandatory gate if UX open)
  • Nyx (build) → Brigid (curl-check + iOS Safari + viewers)
  • Lime (visual) → Brigid (cross-device render)

Acceptance criteria

  • Lighthouse pass + visual diff vs mockup
  • Render OK on real iOS Safari (no simulator)
  • Reduced-motion fallback complete

Validated in

bast-olimpo.pages.dev · perla-encuesta.pages.dev · malabartista.com — Skill /landing-pages

03 Live Design System + Startups that already have product and need to scale UI without hiring an in-house designer.

Output

Figma library + deployable prototype + tokens.json + documented components + internal SOP.

Ariadna's Contract

The DS resolves the 8 recurring bugs already paid in past runs. The prototype is deployable, not decorative.

Verifier loop

  • Lime (tokens) → Brigid (consistency + naming)
  • Nyx (components) → Lime (visual fidelity)
  • Nyx (prototype) → Ariadna (use contract)

Acceptance criteria

  • Prototype opens at public URL without auth
  • Tokens validate against client's brand-guide
  • Zero failures in 8 canonical bugs

Validated in

Brasador DSS · SURI tratamientos — Skill /design-system · SOP SOP_DESIGN_SYSTEM

04 WhatsApp Agent / Conversational Bot + Businesses operating on WhatsApp (clinics, retailers, D2C) that want to automate care/booking/sales without losing brand voice.

Output

Live n8n workflow + Supabase schema + system prompt + tools + protocoled human handoff + E2E tests.

Ariadna's Contract

It's not a bot — it's a conversational agent with brand voice. Sensitive operations require confirmation.

Verifier loop

  • Nyx (schema) → Brigid (post-migration verification)
  • Brigid (workflow) → Nyx (Meta docs cross-check)
  • Ariadna (prompt) → cross-model review (Claude + Gemini + GPT)

Acceptance criteria

  • E2E happy path passes
  • Cross-model returns 0 real GAPs
  • Lock TTL ≤ 90s · dedup active

Validated in

Perla Nanatka v3.0.1 · Conect.ia · Brand Scout — Skill /croquis-agente · SOP SOP_AGENTE_n8n_WHATSAPP

05 Automated Survey / Form + Founders / brands that need to validate product or feedback with data in their own DB and a brand-voiced email — not generic SaaS.

Output

Deployable HTML + SQL migration + 2 JS Code nodes + n8n workflow + deploy plan + humanized email.

Ariadna's Contract

The form doesn't feel like a form — it feels like a conversation with the brand. The follow-up email is editorial, not transactional.

Verifier loop

  • Lime + Nyx (HTML) → Brigid (mobile + SOP gotchas)
  • Nyx (SQL/JS) → Brigid (DB integrity)
  • Sibila + Lime (email) → Ariadna (tone + accents + ñ)

Acceptance criteria

  • Form inserts with all schema fields
  • Email arrives with HTML + plain-text fallback
  • Next survey ships in ≤30 min wall-time

Validated in

Perla beta v02 · Brasador NPS pilot — Skill /encuesta · SOP SOP_ENCUESTA_AUTOMATIZADA_v2

06 Ad Creatives with narrative bible + D2C brands that need creatives for Meta Ads without dropping craft to AI slop, with a narrative bible every ad respects.

Output

Structured JSON brief + ad creative variants (statics + reels) + shot list with 7 narrative beats + sidecar per asset.

Ariadna's Contract

Every ad respects the 7 beats. No AI slop. No recycled hooks.

Verifier loop

  • Scraper (brief) → Ariadna (frame check)
  • Lime + nano-banana → Brigid (real-reference when product is branded)
  • Lime + Nyx (reel) → Ariadna (7 beats compliance)

Acceptance criteria

  • Identifiable 7 beats in pack
  • Branded products use real reference
  • Sidecar complete per asset

Validated in

Brasador (Hola Brasador · Esencial · Vernola) — Skills /scraper · /forge · /ad

Layer 03 · The Infrastructure

Why a deploy ships in minutes.

The cycle isn't theoretical — it runs on layered infrastructure built up run by run. Auth channels, creative pipelines, agent memory, canonical SOPs. Each layer is what turns "spec" into "live URL" in minutes, not weeks.

Deploy & auth channels

always-on credentials
  • Cloudflare Pages with authenticated CLI
  • GitHub CLI with scopes ready
  • Self-hosted n8n with public webhooks
  • Managed databases · multi-environment
  • WhatsApp Cloud API · verified tenant
  • Transactional email · root domain verified
  • Server-side Drive sync
  • Vertex AI access · OpenAI · Anthropic · Gemini in parallel

Creative stack

image · audio · video · OCR
  • AI image generation CLI (Flash + Pro, 1K–4K, real-reference)
  • Local Whisper transcription
  • Video pipelines via ffmpeg
  • Bilingual OCR
  • Color-safe PDF tooling
  • Production-weight image optimization
  • Lime: skill suite + agent memory + outfit system

Olimpo memory layer

how the agents "know"
  • Six specialized agents
  • Twenty-plus invocable skills
  • Indexed global memory · 140+ entries
  • Per-project: CLAUDE / STATE / MEMORIA / HANDOFF
  • Per-agent memory inside each project
  • Append-only activity log + session brief
  • Plan files as durable design artifacts

Canonical SOPs

six packages, six SOPs
  • Image generation (mandatory phases 0+1)
  • Ads narrative bible (7 beats, 18 rules)
  • Design System with Claude Design
  • Survey automation v2 (38 rules + 14 gotchas)
  • WhatsApp agent (12 patterns)
  • Strategic Sprint
  • Parametrized templates per SOP

MCP servers

extended tooling
  • UI component search & examples
  • Browser automation
  • Sequential thinking
  • Design tool integration
  • Calendar & mail productivity
  • Drive document access

Owned servers

always-on infra
  • Primary Linux server with Docker (n8n, bots)
  • Secondary Linux server with Coolify
  • Color-coded terminal profiles per project
  • Subagent invocation by file path
Open for the next run

Let's build what's next.

If a package above maps to something you're trying to ship, we can run it together. The cycle is open.

Get in touch →