Technologies
What runs under our agents.
Six layers, the same in every build. We choose tools for the job, never the other way round.
The stack
Six layers, top to bottom.
- ChannelsWhere people meet the agent.
WhatsApp
Slack
Microsoft Teams
Telegram
- Web chat
- Voice
- Agents and orchestrationWho does which part, and in what order.
LangGraph
CrewAI
- AutoGen
LangChain
MCP
- Tool calling
- ModelsThe cheapest one that passes the tests.
Claude
GPT
Llama
Hugging Face
YOLOv8
Core ML
TensorFlow Lite
- Knowledge and dataWhat the agent is allowed to know.
PostgreSQL
pgvector
Pinecone
LlamaIndexRedis
MongoDB
- Guardrails and testingRules in code, not in the prompt.
- Output checks
- Evaluation suites
- Action logs
Playwright
Lighthouse
- Cloud and deliveryYour cloud, or a private one we run.
AWS
Google Cloud
Azure
Docker
Kubernetes
Terraform
Choosing a model
Four rules before any model is picked.
- 01
Test first
We write the tests a good answer must pass before choosing a model.
- 02
Cheapest that passes
A small model that passes beats a large one that costs ten times more.
- 03
Private when it must be
Sensitive data stays in your cloud, or on the device.
- 04
Swappable
Models sit behind one interface, so a better one is a change, not a rebuild.
Your systems
We connect to what you already run.
- Finance and ERP
TallySAP
NetSuiteQuickBooks
Xero
Zoho Books
- CRM and sales
Salesforce
HubSpot
Zoho CRM
- Support
Zendesk
FreshdeskIntercom
- Commerce and payments
Shopify
WooCommerce
Razorpay
Stripe
- Documents
Google Drive
SharePoint
DocuSignExcel
Google Sheets
- Anything else
- REST APIs
- Webhooks
- Databases
MCP servers
Running something not listed? If it has a screen or an API, we can usually reach it.
Tell us what your team works in every day. We reply within one working day.