Rapi: Personal Intelligence in WhatsApp

The description states that Rapi is a personal intelligence assistant operating within WhatsApp. It performs bounded deep research using tools like Tavily and Exa, verifies claims against fetched sources, and returns concise answers with clickable citations.

It also remembers how clients work — preferred structure, language, depth, recurring interests, and briefing cadence — to improve utility over time.

The product includes:

• A Day 1 engine for bounded research (max 4 sub-questions, max 7 URLs).
• A Day 2 engine that implements the full commercial journey as a resumable state machine.
• A client-facing runtime with no terminal or code execution, keeping all interactions in warm Bahasa Indonesia.

Rapi is built using technologies such as:

• Agent frameworks (Baileys, Hermes)
• LLMs (GPT-5.6, Codex)
• Tools for search and retrieval (Tavily, Exa)
• Infrastructure (Node.js, Python, systemd, Ubuntu)

Inference: The product appears to be a hybrid of AI research assistant and conversational automation, designed for high-trust, low-error environments like senior public figures' workflows.

The description states that Rapi was built because senior decision-makers do not need another dashboard, but rather judgment, continuity, discretion, and evidence in the channel they already use — which is WhatsApp for Indonesia’s senior generation.

It positions itself as:

• A private research assistant in WhatsApp
• An assistant that remembers users and improves over time
• A product that proves its value before any commercial interaction

The claim evolution shows a shift from:

1. Manual setup by one founder → to
2. Fully automated onboarding, research, and provisioning within WhatsApp → to
3. Referral-based growth with real proof-of-concept before payment

Inference: The positioning is rooted in trust, continuity, and channel familiarity — not just technology or features.

The description states that Rapi targets senior Indonesian public figures, including former central bank officials and legal statesmen. These users are described as:

• Having public biographies
• Using WhatsApp as their primary communication channel
• Requiring discretion, continuity, and evidence in their work

It also mentions that the most active user is a former Deputy Governor.

The ICP appears to be:

• High-trust, high-context users (e.g., policymakers, legal experts)
• Decision-makers who prefer WhatsApp over apps
• Users who value memory-aware tools and long-term utility

Inference: The target customer is not a general consumer but a niche group of trusted professionals with specific needs for secure, continuous, and evidence-based intelligence.

The description states that Rapi:

• Works by referral only
• Proves itself with one real research assignment before any payment talk
• Uses a guarded worker to assign pre-paired WhatsApp numbers and provision isolated tenants after payment confirmation
• Has a commercial journey implemented as a resumable state machine

It also says that:

• Users initiate billing conversations
• The system enforces commercial honesty (e.g., no invented prices or destinations)
• It supports onboarding, charging, and provisioning entirely within WhatsApp

However, there is no evidence of pricing models, revenue streams, or actual paid users beyond the 3 production users mentioned.

Inference: The business model seems to be based on referral-driven growth, proof-of-concept before payment, and isolated tenant provisioning, but no concrete commercial details are provided.

The description states that:

• Rapi uses GPT-5.6-sol for all core functions including research planning, claim verification, synthesis, and memory-aware conversation.
• It implements bounded research: at most 4 sub-questions, at most 7 unique URLs.
• The system enforces structural verification — no citation allowed unless source was fetched in that session.
• A replay evaluator scores persisted traces without spending new credits.
• Tools like Codex were used for continuous implementation from requirements to deployment.
• The system includes mechanical research budgets, 10-minute windows, graceful interrupts, and zero-cost replay scoring.

It also mentions:

• No terminal or code execution in client-facing runtime
• All tools, errors, accounts, and quotas are hidden from clients
• Client messages stay in warm Bahasa Indonesia

Inference: The technical stack is designed for security, isolation, and reliability, with strong emphasis on bounded AI behavior and verifiable outputs.

The description states:

• Rapi has served 3 production users since June 2026
• All are senior Indonesian public figures with public biographies
• The most active user is a former Deputy Governor
• A second user, a senior legal statesman, uses it for real work product (annotated book chapters, board analyses)
• Users ask when they can pay — indicating willingness to pay

It also claims:

• Final live research suite passed 5 out of 5 questions
• Day 1 tests 17 out of 17
• Day 2 tests 27 out of 27
• Onboarding conversation fixtures 15 out of 15

However, there is no evidence of revenue, customer acquisition cost, or conversion metrics beyond internal test suites.

Inference: There is limited external traction, but strong internal validation and user engagement among high-value users.

The description does not mention any competitors directly. However, it implies a space where:

• AI assistants are used for research
• Personal intelligence tools exist in various forms (e.g., chatbots, dashboards)
• WhatsApp-based solutions are rare or untested at scale

It positions itself as unique due to:

• Operating inside WhatsApp
• Memory-aware and relationship-based intelligence
• High-trust, low-error design

Inference: Rapi operates in a niche market of high-context, private research assistants — not widely covered by existing solutions.

1. Limited customer base: Only 3 production users; no evidence of scalable demand.
2. Highly specialized use case: Targets only senior public figures, limiting market size.
3. No pricing or monetization data: No evidence of revenue, pricing models, or conversion rates.
4. Self-reported metrics: All performance claims are internal test results — not externally validated.
5. Single-founder team: Team size is 1; no indication of scalability or support infrastructure.
6. WhatsApp-only interface: May limit adoption if users prefer other platforms.
7. No public data or third-party validation: No independent reviews, customer testimonials, or audits.

Inference: The project shows promise in a narrow niche but lacks evidence of broader viability or commercial traction.

1. What is the exact referral process? How are prospects identified and introduced?
2. Can you provide examples of the types of questions clients ask, and how Rapi responds?
3. Are there any plans to expand beyond senior public figures or into other markets?
4. How does Rapi handle edge cases where research fails or sources are unavailable?
5. What is the current cost structure for provisioning isolated tenants?
6. Have you tested with more than 3 users? If not, what’s the plan for scaling?
7. Is there any mechanism to track user retention or long-term usage?
8. How do you ensure data privacy and compliance in a WhatsApp-based system?