Círculo Copy

Private expert skills, exposed only through your agent.

Círculo Copy is a remote skills platform for Hermes users who want to extend their agent with private copywriting, strategy, and feedback capabilities without exposing the underlying playbooks. It lets clients call Rafa’s expertise through a secure connector while keeping the intellectual property and operating rules fully on Rafa’s server. The product matters because it monetizes expertise as an ongoing subscription, protects proprietary methods, and works cleanly with the client’s own credits, documents, and agent setup.

Business Goals

  • Reach 25 paying teams within 90 days of launch with at least 60% connecting an active Hermes agent.
  • Achieve a 35% trial-to-paid conversion rate by month 4 through a 7-day guided activation flow.
  • Keep monthly gross margin above 70% by limiting AI compute usage with quotas and tiered plans.
  • Drive under 5% monthly churn among active customers after the first 3 billing cycles.
  • Generate at least 20% of new revenue from referrals or partner-led sales by month 6.

User Goals

  • Let customers use Rafa’s private skills from their own Hermes agent without seeing the source methods.
  • Make setup simple enough that a non-technical operator can connect in under 15 minutes.
  • Return useful copy, critique, and strategy outputs inside the customer’s own workflow.
  • Allow customers to keep their own documents, memory, and custom skills alongside the connector.
  • Ensure access stops immediately when a subscription expires or payment fails.

Non-Goals

  • Do not ship downloadable or installable skill files that expose Rafa’s private prompts or frameworks.
  • Do not replace the customer’s Hermes instance or manage their core agent identity.
  • Do not build a general marketplace for third-party skills in the first release.
  • Do not support unrestricted BYOK key management in the MVP.

Agency Owner Laura, 38 - Laura runs a small performance marketing agency and uses Hermes to help her team draft emails, sales pages, and client feedback. She wants to add a premium copy system without hiring another specialist or exposing the method to clients.

Agency Owner Laura, 38

  • As a agency owner, I want my Hermes agent to call Círculo Copy securely, so that my team can use the method in daily work.
  • As a agency owner, I want billing to stop access automatically when a plan is paused, so that I can control cost and usage.
  • As a agency owner, I want to see usage by client or workspace, so that I can allocate spend internally.

Freelance Consultant Diego, 44 - Diego sells strategy and messaging services to founders and needs a way to scale his process without packaging it as a visible product. He wants a private service that acts like his brain, not a public software course.

Freelance Consultant Diego, 44

  • As a consultant, I want to expose only a connector to clients, so that my method stays hidden.
  • As a consultant, I want my own internal prompt rules to stay on my server, so that clients cannot copy my framework.
  • As a consultant, I want simple logs of what was requested and returned, so that I can improve the service over time.

Ops Admin Marta, 29 - Marta manages access, invoices, and integrations for a small subscription-based knowledge service. She needs reliable authentication, clear status, and the ability to deactivate users quickly.

Ops Admin Marta, 29

  • As an operations admin, I want to issue and revoke workspace access quickly, so that billing and entitlements stay accurate.
  • As an operations admin, I want to monitor failures and latency, so that I can catch connector issues before customers notice.
  • As an operations admin, I want clear audit trails, so that I can resolve disputes and support requests.

Secure Remote Skill Connector · High priority

  • Provide a Hermes-compatible connector that lets the customer agent call Rafa’s private skills through a remote API or MCP server without exposing source logic.
  • Support authenticated tool calls from Hermes using API keys or OAuth-style workspace tokens.
  • Return structured responses that Hermes can consume directly, with optional human-readable summaries.
  • Reject unauthenticated, expired, or over-quota requests with clear error messages.
  • Support versioned skill endpoints such as copy analysis, email rewrite, sales page feedback, and voice calibration.
  • Log request metadata without storing unnecessary private content by default.

Subscription and Entitlement Control · High priority

  • Control access based on active subscription status, usage limits, and workspace entitlements so the service can be monetized predictably.
  • Check subscription status on every request with a cached entitlement layer for low latency.
  • Support plans with monthly request quotas, soft warnings, and hard cutoffs.
  • Automatically disable access on failed payment, cancellation, or expired trial.
  • Allow admin overrides for grace periods and manual comp access.
  • Show current plan, remaining quota, and renewal date in the dashboard.

Private Knowledge Execution · High priority

  • Execute Rafa’s hidden methods on the server side so clients receive outputs without seeing prompts, frameworks, or internal rules.
  • Keep master prompts, method instructions, and evaluation rubrics server-side only.
  • Use prompt templates that can be updated without changing the client connector.
  • Support deterministic output schemas for analysis, rewrite, critique, and suggestions.
  • Allow per-skill guardrails such as tone, audience, and forbidden claims.
  • Store skill version history for rollback and A/B testing.

Workspace Dashboard · Medium priority

  • Give customers and admins a simple web dashboard to manage connections, usage, billing, and logs.
  • Display connected Hermes workspaces, status, and last activity.
  • Show monthly usage by connector type and request volume.
  • Provide quick links to rotate keys, pause access, and view invoices.
  • Allow filtering logs by date, skill type, and error state.
  • Include onboarding checklists and setup health indicators.

Observability and Audit · Medium priority

  • Track reliability, billing events, and security-relevant actions to support customer trust and troubleshooting.
  • Record every tool call with timestamp, workspace ID, skill name, latency, and outcome.
  • Keep audit logs for auth changes, plan changes, and revocations.
  • Expose operational metrics such as error rate, timeout rate, and token spend.
  • Alert on abnormal spikes, repeated failures, or suspicious request patterns.
  • Support export of logs in CSV and JSON for support and compliance.

Fast Setup for Hermes Users

  • User signs up for Círculo Copy and chooses a plan or starts a trial.
  • User connects a Hermes workspace with a secure token or connector URL.
  • User selects the first enabled skill pack such as copy review, email rewrite, or feedback.
  • User runs a test request from Hermes and sees a successful response in under 15 minutes.
  • User reaches first value by receiving a usable rewrite, analysis, or critique inside their own agent.
  • If setup fails, the dashboard highlights the exact cause: expired token, blocked workspace, or quota limit.

1. Connect Workspace

  • The user links their Hermes agent to the Círculo Copy service with a secure credential and assigns a workspace name.
  • Validate token format and workspace ownership before enabling access.
  • Show a clear success state with the connector endpoint and status.
  • If auth fails, provide a retry path and a plain-language fix.

2. Choose Skills

  • The user activates one or more remote skills based on their plan and use case.
  • Display skill descriptions without revealing internal method details.
  • Prevent enabling skills that are not included in the plan.
  • Let admins add or remove skills per workspace.

3. Send a Request

  • Hermes sends a structured prompt or tool call to the remote service and receives a hidden-method response.
  • Support common payloads like email text, landing page copy, or campaign brief.
  • Reject malformed payloads with schema validation errors.
  • Timeout gracefully and return a recoverable retry message.

4. Get the Output

  • The service returns a final answer that Hermes can present to the user or continue processing with.
  • Return the output in a consistent schema with summary, recommendations, and revised copy.
  • Support both short answers and long-form structured critiques.
  • Include optional confidence or priority labels where useful.

5. Monitor and Improve

  • The user reviews usage, failures, and performance from the dashboard and adjusts plan or skill mix.
  • Surface quota usage before a hard cutoff occurs.
  • Show recent requests, latency, and failure reasons.
  • Allow easy plan upgrade from within the dashboard.

Advanced Controls and Edge Handling

  • Per-skill rate limits and per-workspace quotas.
  • Prompt versioning with rollback and canary release support.
  • Workspace-level custom tone settings and audience presets.
  • Manual approval mode for high-risk outputs such as claims or legal-adjacent text.
  • Webhook notifications for payment failure, access revocation, and connector errors.

Trustworthy, Fast, and Clear UI

  • Minimal dashboard with clear access status, remaining quota, and recent activity.
  • Prominent privacy messaging that states methods stay on Rafa’s server.
  • Accessible design with high-contrast states, keyboard navigation, and screen-reader labels.
  • Fast-loading pages with skeleton states and immediate connector feedback.
  • Error states written in plain language, with one-click remediation steps.

Laura runs a small agency and uses Hermes every day to draft client emails and sales pages. She wants to add a stronger copy framework, but she cannot afford to expose her process or let clients inspect the method behind it.

She connects her Hermes workspace to Círculo Copy, activates email rewrite and sales-page critique, and tests a request in less than 15 minutes. Hermes calls Rafa’s remote skill server, gets a polished response, and Laura’s team keeps working inside the same agent they already use.

A month later, Laura has faster output, better consistency, and no leaked playbooks. Rafa has a repeatable subscription business, usage visibility, and a product that stops working the moment the plan ends, which makes the intellectual property actually monetizable.

User-Centric Metrics

  • Average first successful connector setup under 15 minutes for 80% of new workspaces.
  • At least 70% of trial users send a second request within 24 hours.
  • Median remote skill response time below 2.5 seconds for standard text tasks.
  • Customer-reported output usefulness rating of 4.5/5 or higher.
  • Fewer than 3% of requests require manual support intervention.
  • At least 60% of active workspaces use the service weekly by month 2.

Business Metrics

  • Trial-to-paid conversion of 35% by month 4.
  • Monthly churn below 5% after the first full billing cycle.
  • Net revenue retention above 110% through plan upgrades and higher usage tiers.
  • At least 25 active paying workspaces within 90 days of launch.
  • At least 20% of new customers acquired through referrals or partner channels by month 6.

Technical Metrics

  • Service uptime of 99.9% monthly.
  • P95 request latency under 3 seconds for standard skills.
  • Authentication and entitlement checks complete in under 200 ms at P95.
  • Zero critical findings for exposed prompt or secret leakage in internal security reviews.

Tracking Plan

  • Track signup_completed when a user creates an account.
  • Track workspace_connected when Hermes credentials are successfully verified.
  • Track skill_enabled when a skill pack is activated.
  • Track skill_request_sent with workspace, skill, and payload category.
  • Track skill_response_success with latency, status, and output type.
  • Track payment_failed and subscription_canceled for churn analysis.
  • Track quota_warning_shown and quota_limit_reached to monitor monetization pressure.

Technical Needs

  • Backend API using Node.js with Fastify or NestJS for authenticated skill routing.
  • Remote tool layer implemented as an MCP-compatible server or REST API gateway.
  • PostgreSQL for subscriptions, workspaces, skill configs, and audit records.
  • Redis for short-lived entitlement caching, rate limits, and session tokens.
  • Background jobs with BullMQ or Temporal for billing sync, retries, and alerts.
  • Frontend dashboard built with Next.js and a component library such as shadcn/ui.
  • Observability with OpenTelemetry, Sentry, and structured JSON logs.

Integration Points

  • Hermes agent connector via MCP or API endpoint.
  • Stripe for subscriptions, invoicing, failed payment handling, and upgrades.
  • Email delivery and notifications via SendGrid or Resend.
  • Authentication via magic links or Google OAuth for the dashboard.
  • Optional webhook integration for CRM or Slack notifications on usage and billing events.

Data Storage & Privacy

  • Store only the minimum request payload needed to process the skill; redact sensitive text in logs by default.
  • Keep master prompts, internal rubrics, and method documents in a private server-side repository with restricted access.
  • Support GDPR deletion requests for user accounts, workspaces, and logs within defined retention windows.
  • Encrypt data at rest with AES-256 and in transit with TLS 1.2 or higher.
  • Separate customer workspace data from Rafa’s private skill assets at both application and storage layers.

Scalability & Performance

  • Cache entitlement and plan lookups to avoid billing-service bottlenecks.
  • Use async job queues for long-running or high-token tasks.
  • Scale the skill execution layer horizontally behind a load balancer.
  • Set per-workspace request limits to prevent abuse and cost spikes.

Potential Challenges

  • Risk: clients may try to infer hidden methods from outputs. Mitigation: use output shaping, redaction rules, and limited surface area per skill.
  • Risk: latency could feel slow if every request calls remote processing. Mitigation: cache entitlements, optimize prompts, and keep standard responses under 3 seconds P95.
  • Risk: payment failures could create support friction. Mitigation: add grace periods, clear warnings, and self-serve plan recovery.
  • Risk: connector misconfiguration may block adoption. Mitigation: provide guided setup, validation checks, and one-click test requests.
  • Risk: privacy expectations may be unclear. Mitigation: publish a plain-language data policy and visible privacy indicators in the dashboard.

Team & resourcing - Small team - 2 engineers, 1 product designer, part-time PM, and fractional support/ops.

Phase 1: MVP Connector · Weeks 1-4

  • Hermes-compatible authenticated connector
  • One private skill pack for email rewrite and copy critique
  • Basic subscription gating with trial and paid plan
  • Minimal dashboard with connect, test, and usage views
  • Stripe billing integration and access revocation

Phase 2: Private Skill Platform · Weeks 5-8

  • Skill versioning and rollback
  • Structured response schemas for multiple skill types
  • Usage quotas, warnings, and plan upgrades
  • Audit logs and admin controls
  • Improved onboarding and guided setup

Phase 3: Reliability and Trust · Weeks 9-12

  • Monitoring, alerts, and support tooling
  • Performance tuning to hit latency and uptime targets
  • Privacy and retention controls with delete/export flows
  • Expanded skill packs such as sales pages, feedback, and voice calibration
  • Webhook and notification integrations

Phase 4: Growth and Optimization · Weeks 13-16

  • Referral and partner onboarding
  • A/B testing for prompts and response formats
  • Advanced analytics dashboard
  • Enterprise-ready workspace controls
  • Canary releases for new skill versions

Paste this into Cursor, Bolt, Lovable, or v0 to start building.

Build a SaaS product called Círculo Copy: a private remote skills platform for Hermes users.

Goal: let a customer connect their Hermes agent to Rafa’s hidden copywriting skills through a secure MCP-compatible server or REST API, without exposing the internal prompts, frameworks, or source methods. Customers pay a subscription in Stripe, connect a Hermes workspace, enable skills, and then Hermes calls the remote service for outputs like email rewrites, copy critique, sales-page feedback, and voice calibration.

Core requirements:
- Public marketing/signup pages
- Authenticated dashboard with workspace connection status, plan, quota, recent requests, billing, and logs
- Secure connector endpoint for Hermes tool calls
- Hidden server-side skill execution with versioned skills and structured JSON responses
- Stripe subscriptions with trial, active, past_due, canceled states and automatic entitlement enforcement
- Admin controls to pause/revoke access, rotate tokens, and inspect audit logs

Primary screens/flows:
1) Landing page with value prop, privacy promise, and CTA
2) Signup and login
3) Connect Hermes workspace flow with token validation and test request
4) Skill selection screen
5) Usage and billing dashboard
6) Admin logs and access control screen

Data model:
- User: id, name, email, role
- Workspace: id, userId, name, hermesConnectorId, status, quotaLimit, quotaUsed
- Subscription: id, workspaceId, stripeCustomerId, plan, status, renewalDate
- Skill: id, name, description, version, enabled
- SkillRequest: id, workspaceId, skillId, inputType, status, latencyMs, createdAt
- AuditLog: id, actorId, action, targetType, targetId, metadata, createdAt

Tech stack default:
- Next.js 14 with App Router, TypeScript, Tailwind, shadcn/ui
- Node.js backend using NestJS or Fastify
- PostgreSQL with Prisma
- Redis for caching and rate limits
- Stripe for billing
- Sentry and OpenTelemetry for monitoring
- Deploy on Vercel for frontend and Fly.io or Render for API

Build it as a production-ready MVP with secure auth, clean UI, clear error states, and placeholder data where needed. Use plain, professional copy in Spanish or English depending on the app surface, but keep the architecture and code ready for multi-language support.

Business Idea

igual no me he explicado bien. Ellos montan su agente en Hermes y yo soy como un suministrador de habilidades que no pueden ver pero SÍ usar mientras estén pagando. Como es su bot, pueden sumarle los concimeintos/ skills suyos que quieran que para eso es suyo... pero los créditos, etc... los pagan ellos Pensó durante 39s Ahora sí. Y la respuesta cambia un poco: Sí, es viable como modelo de negocio. Pero no sería “darles skills de Hermes” en sentido literal. Sería más bien: Ellos tienen su Hermes. Ellos pagan sus créditos. Ellos añaden sus documentos, memoria y skills. Y tú les das acceso a una capa remota de habilidades Rafa que su agente puede usar, pero no leer. La diferencia es clave. Lo que NO sirve Esto no te conviene: Socio instala Hermes ↓ Instala "skills de Rafa" ↓ Los archivos quedan en su ordenador ↓ Los puede ver, copiar o seguir usando En Hermes, las skills normales viven como archivos en la carpeta de skills del perfil. La documentación dice que están en ~/.hermes/skills/, que aparecen en el índice de skills y que incluso pueden verse/gestionarse desde el sistema de skills. Además, si compartes un perfil completo como “distribution”, Hermes lo plantea como un repositorio Git con SOUL.md, config.yaml, skills/, etc.; los usuarios pueden instalarlo, actualizarlo, conservar sus propias memorias y claves, pero también pueden inspeccionar el repo. Así que si tus skills llegan físicamente a su Hermes, ya no son realmente secretas. Lo que SÍ serviría El modelo bueno sería este: Hermes del socio ├── sus créditos/API key ├── sus documentos ├── sus skills propias ├── su memoria └── conector a "Skills Rafa" Skills Rafa └── viven en TU servidor / servicio remoto ├── copywriting ├── emails ├── cartas de venta ├── feedback ├── voz ├── estrategia └── método Círculo Copy El socio no instala tus skills. Instala un conector. Algo tipo: /circulo-copy analiza este email o: Usa las habilidades de Rafa para mejorar esta secuencia. Y su Hermes llama a tu servicio. Tu servicio procesa la petición usando tus skills privadas y devuelve la respuesta. El socio ve esto: Aquí tienes el análisis. Aquí están los problemas. Aquí va la versión corregida. Pero no ve esto: Las instrucciones internas. Los frameworks completos. Los criterios exactos. Los archivos fuente. Los prompts maestros. La pieza técnica más lógica Para esto, lo más limpio sería un MCP server remoto o una API propia. Hermes soporta MCP para conectar el agente a herramientas externas, incluyendo servidores remotos HTTP, APIs internas y herramientas que viven fuera de Hermes. La documentación lo describe precisamente como una capa para que Hermes use herramientas externas sin que esas herramientas estén dentro del propio Hermes. La arquitectura sería: Hermes del socio ↓ MCP / API "Círculo Copy Skills" ↓ Servidor tuyo ↓ Skills privadas de Rafa ↓ Respuesta al Hermes del socio Y ahí sí puedes controlar el pago: Socio activo → acceso permitido Socio deja de pagar → acceso denegado Sobre los créditos Aquí hay un matiz importante. Si ellos pagan sus créditos, su Hermes puede pagar: La conversación general El uso de su propio agente Sus documentos Sus propias skills Pero cuando llaman a tus skills remotas, hay varias opciones. Opción 1 — Ellos pagan sus créditos, tú solo das lógica Tu servicio devuelve reglas, criterios, análisis estructurado o instrucciones parciales, y el Hermes del socio usa su propio modelo para redactar. Ventaja: El coste principal lo pagan ellos. Problema: Si devuelves demasiada lógica, puedes revelar parte de tus skills. Opción 2 — Tú haces el procesamiento fuerte en tu servidor El socio envía el texto. Tu servidor usa tus skills y un modelo. Devuelves el resultado final. Ventaja: Tus skills quedan mucho más protegidas. Problema: El coste de modelo lo pagas tú, salvo que lo repercutas en la cuota. Opción 3 — BYOK: Bring Your Own Key El socio mete su propia API key en tu sistema remoto. Tu servidor usa tus skills privadas, pero factura contra la key del socio. Ventaja: Ellos pagan los créditos. Tú proteges las skills. Problema: Es más delicado a nivel técnico, privacidad y confianza. Tendrías que gestionar claves de terceros con mucho cuidado. Para empezar, yo no haría BYOK. Haría algo más sencillo: Cuota de Círculo Copy + uso razonable incluido + límite mensual + si alguien consume mucho, plan superior Entonces, ¿se puede hacer tal como lo imaginas? Sí, pero no como “skill instalada e invisible”. Eso, en local, no lo veo seguro. La forma correcta sería: Skill visible del socio: "Conector Círculo Copy" Skill invisible de Rafa: vive en tu servidor Relación: el conector llama a tu servidor Algo así: Hermes del socio ├── Skills propias del socio ├── Conocimiento propio del socio ├── Créditos/API propios └── Skill/conector Círculo Copy ↓ autenticación ↓ servidor Rafa ↓ skills privadas ↓ respuesta La frase de negocio sería muy buena No vendes: “Te doy mis skills.” Vendes: “Tu Hermes puede usar las habilidades de Círculo Copy mientras seas socio.” Eso es mucho más fuerte. Porque el socio conserva su agente. Conserva sus documentos. Conserva sus créditos. Conserva sus skills. Pero pierde acceso a tu capa diferencial si deja de pagar. Conclusión La idea es viable. Pero la regla técnica sería esta: Si la skill vive en su ordenador, no está protegida. Si la skill vive en tu servidor y su Hermes solo la llama, sí puedes protegerla. Para Círculo Copy, el producto ideal sería: Círculo Copy Skill Server o, más comercial: Motor Rafa para Hermes Los socios instalan un conector. Tú controlas licencias. Ellos pagan su uso normal. Y tus habilidades no salen de tu casa.

Make My PRD

Design by The Resonance | Powered by GPC – The AI Transformation Company

    PRD: igual no me he explicado bien