Product Architecture
The ind.ai Starter Agent Fabric
Free starter agents for 1M+ users — without 1M autonomous runtimes
ind.ai can scale by giving every user an agent identity, compact memory, and access to a shared orchestration fabric — instead of spinning up a dedicated always-on agent for each person.
This architecture allows ind.ai to offer mass-market starter agents at low cost, while reserving deeper autonomy, persistent workflows, and private runtimes for premium or enterprise tiers.
Core Model
The unit of scale
1 User
=
Users should experience ind.ai as "my personal agent system," but the backend should operate as a multi-tenant shared agent fabric — not as millions of separate agent processes.
Strategic Context
Why this model is necessary
This is a scaling choice, not a compromise.
Infrastructure cost explodes
Giving every free user a persistent autonomous runtime is economically unviable at India scale. Shared orchestration keeps unit costs flat.
Orchestration complexity compounds
Millions of independent agent processes create an unmanageable coordination surface. A central router with controlled tool access stays governable.
Monitoring and abuse surface grows
Autonomous agents without centralised policy enforcement become a safety and compliance risk. Shared fabric means shared guardrails.
Free-tier economics break
Without this architecture, free-tier agents become unsustainably expensive — forcing either aggressive rate limits or premature monetisation walls.
Autonomy is wasted on lightweight tasks
Most starter-agent interactions are simple: search, draft, remind, navigate. Full autonomy is unnecessary — and expensive — for these workflows.
Architecture
Layered architecture stack
Seven layers from identity to premium runtimes. The shared orchestration core is what makes the free tier economically viable.
Identity Layer
User handle, language, profile, permissions, consent, and account tier.
Session Layer
Ephemeral sessions spin up only when the user sends a request. No always-on runtime for free users.
Memory Layer
Compact memory store with profile facts, preferences, summaries, saved items, connector permissions, and retrieval-friendly context.
Orchestration Layer
Central router decides intent, selects the right starter-agent mode, chooses allowed tools, and determines model cost tier.
Tool Layer
Shared tool APIs for search, drafting, reminders, service navigation, forms, payments, commerce, and verified workflows.
Worker / Queue Layer
Longer or slower jobs run asynchronously through workers rather than blocking the live conversation.
Premium Runtime Layer
Persistent, more autonomous, or sandboxed runtimes reserved for paid, developer, or enterprise tiers.
Designed for 1M+ identities, not 1M active runtimes.
Starter Agents
Starter agents for free users
Each card represents a product surface — not a separate always-on runtime.
Primary Agent
General-purpose assistant that knows the user profile and routes tasks.
Search & Explain
Finds, summarises, compares, translates, and explains information.
Drafting & Forms
Helps write messages, applications, forms, and structured submissions.
Tasks & Reminders
Handles lightweight planning, reminders, follow-ups, and simple scheduling.
Services Navigator
Helps users discover the right service, workflow, next step, or provider.
Commerce & Payments
Assists with transactions, payment-related flows, and trusted service actions with explicit approval.
These are mostly modes + permissions + workflow packs, not separate dedicated agent runtimes.
All starter agents are built using the ind.ai Agent OS architecture.
Request Flow
How a request flows
Example: "Help me book a fertility appointment"
User sends request
Identity + permissions loaded
Router detects intent
Starter-agent mode selected
Memory retrieved
Model tier chosen
Tool / workflow called
Response returned
Long-running → worker queue
Approval before irreversible action
User sends request
Identity + permissions loaded
Router detects intent
Starter-agent mode selected
Memory retrieved
Model tier chosen
Tool / workflow called
Response returned
Long-running → worker queue
Approval before irreversible action
Tier Boundaries
Free vs Pro vs Enterprise
The tier model makes monetisation logic transparent. Free users get useful agents; paying users get depth, persistence, and autonomy.
Free
- Shared orchestration layer
- Ephemeral sessions
- Compact memory
- Limited daily usage
- Basic starter agents
- Restricted tools
- Approval-gated actions
- No persistent autonomous runtime
Pro
- Larger memory
- More usage
- Deeper workflows
- More connectors
- Richer agent customisation
- Limited persistent tasks
- Premium model escalation
- Optional private runtime features
Enterprise
- Dedicated isolation
- Stronger governance controls
- Deeper workflow automation
- Team / org policies
- Auditability
- Private connectors
- Advanced autonomy under guardrails
- Possible OpenClaw-style runtime
Strategic Note
Where OpenClaw fits
OpenClaw is not the default runtime for the ind.ai free tier.
OpenClaw-style runtimes are powerful for developers, power users, and private-agent environments — but not the right economic base layer for a mass-market free tier. For ind.ai, they fit better as:
Cost Discipline
Cost-control principles
Operational discipline that makes the free tier sustainable.
Strategic Advantage
Why this architecture wins
Scalable free-user growth
Millions of identities without proportional infra cost.
Lower cost per user
Shared fabric instead of dedicated runtimes.
Clean monetisation path
Natural progression from free → pro → enterprise.
Better governance & safety
Centralised policies, consent, and audit.
Easier product iteration
One orchestration surface to update, not millions.
Premium upsell path
Deeper autonomy and persistence for paying users.
The right way to scale ind.ai is to productise personal agents — not provision millions of runtimes.
ind.ai should make every user feel like they have a personal AI system, while the backend runs as a shared, policy-driven, multi-tenant agent fabric. That is the architecture that can support India-scale adoption.
Related in India AI