Welcome to session 46! This week Swapnika Nag walks through how Perisclaw is building a WhatsApp-native executive assistant and what it takes to make agents work inside the channel where much of the non-US world already operates.
Swapnika is the Founder of Periskope and Perisclaw. Her team has spent years building WhatsApp operations infrastructure for companies across India, LATAM, the Middle East, Southeast Asia, Europe, and other markets where WhatsApp is core business infrastructure.
Session Overview
This session explains why WhatsApp-native agents are different from generic chatbots, why Perisclaw works through your own number instead of an external bot, and what a production-grade agent harness needs. Swapnika covers channel infrastructure, context assembly, runtime loops, session compaction, permissions, secure containers, memory, jobs, triggers, skills, telemetry, and why the harness itself becomes the product.
Key Takeaways
- WhatsApp is the operating layer for much of the world: In many markets, personal and business coordination already happens on WhatsApp, so agents need to live there instead of forcing users into a new app.
- A useful WhatsApp agent is not a chatbot: It needs context, judgment, scoped permissions, autonomous jobs, and the ability to act through the user's own number.
- The harness matters more than the model: Channel reliability, context assembly, runtime control, memory, permissions, telemetry, and cost control are what make the agent usable in production.
- Security has to be designed into the workflow: Perisclaw uses scoped access, read/write boundaries, isolated encrypted containers, and user-controlled permissions.
- The demo shows the real workflow: Swapnika uses Perisclaw to catch up from self-chat, summarize WhatsApp activity, invoke the agent in groups, coordinate meeting times, and set up access with an invite code.
Topics Covered
Why WhatsApp-Native Agents
Swapnika starts with the Periskope thesis: operations for many companies run on WhatsApp, not just marketing. For consumers and prosumers, a second app does not work; the agent has to live where the work already happens.
Perisclaw Product Decisions
Perisclaw lives in your self-chat and works through your own number. It can be invoked in other chats, given scoped access, and asked to track commitments, reminders, scheduled messages, routines, and autonomous coordination jobs.
Anatomy of the Agent Harness
The session breaks down the major components of a WhatsApp-native agent: the channel, context assembly, runtime loop, session management, permissions, secure runtime, memory, jobs, triggers, skills, logging, and telemetry.
Live Demo
Swapnika shows her personal WhatsApp workflow: morning catchups, to-dos, waiting-on lists, group summaries, group invocation, autonomous meeting coordination, and first-time Perisclaw setup.
Q&A Highlights
- Enterprise calendar access and Google Workspace support
- How conversations are tied back to contacts and threads
- WhatsApp integration risk, unofficial APIs, and Meta policy changes
- Number blocking risk and why spam behavior matters most
- Data residency, DPDP/GDPR considerations, and secure containers
- Building on Periskope APIs and MCP connectors
- The APPLAI invite code shared during the session
Here's the entire recording of the session.