Multi-tenant with strict isolation
Each customer's data, configuration and logs are kept strictly separate.
Defragment Platform · In build
The governed foundation under every agent.
One platform runs every solution. It receives messages from any channel, routes each request the fastest safe way, calls your systems through controlled connectors, runs calculations in rule engines, asks people for approval, and records everything.
Architecture
Six layers, one governed runtime.
A request travels top to bottom. Every hop is permission-checked and written to the audit trail.
- L1
Channels
WhatsApp Business, web chat and email adapters normalise every inbound message into one event format. Voice is on the roadmap.
- Web chat
- Voice (later)
- L2
Router and model gateway
Each request is classified and routed the fastest safe way — a deterministic path where one exists, a model call only where language is needed. One gateway fronts leading commercial and open models and routes by cost, region or your policy.
- Intent routing
- Model gateway
- Per-team budgets
- Region pinning
- L3
Agents
Domain agents are composed from tested, versioned components. Each agent has a declared scope, a set of tools it may call and the approvals it must request.
- Declared scope
- Tool allow-lists
- Versioned flows
- L4
Rules and calculations
Commissions, eligibility, limits and attribution run in rule and calculation engines you can review and test. The model explains results; it does not compute them.
- Rule engine
- Calculation engine
- Test fixtures
- L5
Connectors
Controlled connectors to CRM, ERP, data platforms and document stores. Every call is permission-checked per tenant, per agent and per action.
- CRM
- ERP
- Data platforms
- Documents
- L6
Approvals and audit
Sensitive actions pause for a named approver. Every message, decision, approval and system call is logged with who, what and when.
- Human approvals
- Audit trail
- PII masking
Built in
Controls that come with every agent.
Cost and usage controls
Budgets and usage limits per team and per agent, with model routing that respects them.
Tested before release
Every change runs against your scenarios before it goes live, and can be rolled back in minutes.
No training on your data
Your data is never used to train AI models.
Deployment
Managed service or your own cloud.
Run Defragment on our managed service, in the region you choose, or deploy it inside your own cloud account. Azure is first; other clouds follow.
Option 1
Managed service
We operate the platform for you in the region you choose.
Option 2
Your cloud
The platform runs in your own cloud subscription, under your network and identity controls. Azure first.
Want the technical deep-dive?
We share architecture notes and our current security pack with technology and risk teams.
kumara@agniailabs.in