AI Platform Brief¶

Goal¶

Design and launch a multi-tenant AI service platform for Souza Hub, currently aligned with the Tentacles AI segment, that can support multiple clients, multiple agents per client, and multiple channels over time.

Product Direction¶

The platform should start simple for clients and become more capable over time without locking the business into a single model provider, channel, or infrastructure pattern.

Initial Business Objective¶

Ship a useful working prototype quickly.

The first version should favor:

client simplicity
clear value
predictable pricing
safe staged rollout

Desired Capabilities¶

client-facing AI agents
WhatsApp interaction through 0480 804 100
email-based interaction and training
future website integration
future website generation capability
future voice and telephony integration
support for multiple virtual employees per client

Examples of virtual roles:

customer service
social media assistant
email/admin assistant
role-specific specialist agent

Commercial Principles¶

the product must work before it becomes broad
client experience should feel simple and safe
pricing should be understandable
architecture should remain flexible
future owned-infrastructure inference is desirable, but not required for v1

Technical Direction¶

Initial planning direction from the handoff favors:

Dify
n8n
Langfuse
PostgreSQL
Qdrant or similar retrieval backend
hosted model APIs first

V1 Channel Decision¶

WhatsApp is mandatory for v1
email should also be supported in v1 if it does not materially delay the first working prototype
web chat and voice should remain later-stage channels

Constraint¶

Do not assume ChatGPT Plus or Pro subscriptions are the technical backend for client workloads.

If clients are eventually paying for model usage directly, the cleaner long-term model is:

client-owned API keys
or a managed shared-platform model with clear billing boundaries

Working Rule¶

Prototype first.

Do not overbuild the control plane before there is a validated client workflow.