AI stack
Five layers. Governance on top.
For teams adding AI to Salesforce: models, agents, memory and automation chosen layer by layer, with governance on top. Send the layer that hurts and hear back within 24 hours.
- 01Models · Claude, Gemini, OpenAI, Bedrock
- 02Agentic systems · Agentforce, CrewAI, ADK, A2A
- 03Agentic memory · Neo4j, Cognee, IndyKite
- 04Automation and voice · n8n, Retell
- 05Deployment and governance · identity, evals, policy, audit
Layers
From models up to governance.
Start with the layer your process needs. Compare the tools, see where they fit, and ask what must work before you commit.
Models
The reasoning layer. We pick the model that fits the job, then deploy it so data stays where it should.
Agentic systems
How agents plan, call tools, and talk to each other. Frameworks and protocols, not a single vendor lock-in.
Agentic memory
What the agent remembers: graphs, identity, and long-term context so answers stay grounded.
Automation and voice
Workflows and voice agents that connect models to Salesforce and the rest of the stack.
Deployment and governance
The work that turns a pilot into a production system: identity, evaluations, policy and audit.
One request
How one request moves through the stack.
A customer asks a question on WhatsApp. Five stops later the record is right and someone can see why.
Surface
On WhatsApp, Slack, a portal or the phone. The agent acknowledges at once.
Gateway
The request runs as that person. The gateway decides which tools and models it may call.
Runtime
Agentforce if the work lives in Salesforce. A code agent if it does not.
Memory
The customer profile, the last conversation and the approved articles, with permissions attached.
Write
A queued write with retries. A timeout never posts twice. The trace is kept.
Which layer hurts
Three desks, three missing layers.
Typical first conversations, each starting from one layer.
Example
CTO, Riyadh consumer lender
Today: Three teams chose three models from three demos. Nobody can say where customer data is processed.
After: The candidates run the same set of real cases, and the hosting region is named in writing before the pilot.
For CTOsExample
Head of service, Dubai property developer
Today: The assistant answers buyers well, then forgets what it promised them last week.
After: Profile, working context and past decisions are kept as memory, with the record’s permissions attached.
Real estateExample
Operations lead, Kochi NBFC
Today: Reminder calls end in a notebook. Salesforce never hears what the borrower said.
After: A voice workflow writes each outcome to the loan record and hands the unsure calls to a person.
Automation and voice
Tools
What we actually put in.
The honest no: if the work is a software factory with a named delivery date, we are the wrong partner.
Claude (Anthropic)
Deploy Claude enterprise-wide via API, AWS Bedrock, or Azure.
Google Gemini
Gemini for Workspace and Vertex AI, implemented and governed.
Jev (TypeSafe)
System One model for the small decisions. Pairs with the model that writes.
Sarvam AI
Hears and reads 22 Indian languages, speaks 10. Written back to Salesforce.
OpenAI
GPT models and the Agents SDK. Azure OpenAI when you need it.
AWS Bedrock
Private multi-model AI inside your VPC. AWS HIPAA BAAs and SOC 2 available.
Google ADK
Google Agent Development Kit on Vertex AI.
A2A Protocol
Agent2Agent interop across CrewAI, ADK, and Agentforce.
Claude Agent SDK
Multi-turn Claude agents that validate output before writing.
OpenAI Agents SDK
Typed agent steps: structured outputs and function calling.
LangGraph
Cyclic agent graphs with human approval gates.
Agentforce
Salesforce-native agents. Use when the work lives in the org.
CrewAI
Multi-agent automation that replaces brittle RPA scripts.
MuleSoft Agent Fabric
Orchestrate AI agents across your Salesforce ecosystem.
Model Context Protocol
Standard tool and data connectors for AI agents.
Neo4j
Graph memory and GraphRAG for enterprise agents.
Cognee
Agentic memory so agents remember across sessions.
IndyKite
Identity knowledge graph for agent authorization.
n8n
AI workflow automation connecting Salesforce, AI, and your tools.
Retell AI
Voice AI agents for customer service and sales.
Enterprise AI Deployment
From pilot to production: strategy, governance, and adoption.
AI Governance
EU AI Act, DIFC Reg 10, NIST AI RMF compliance frameworks.
Architecture
Layers are parts. Planes make one system.
The stack is what we install in your org. The planes are how it is assembled, so an agent does not end up as a chatbot with write access.
Digital labor architecture
Five planes, five axioms, and the first slice.
Experience plane
Slack, portals, and voice. Where people meet the work.
Control plane
One front door for every tool call and model call.
Runtime plane
Agentforce, code agents, and the SDKs.
Memory plane
Truth, working context, and decision history.
Durable plane
Queues, retries, and the write that has to hold.
The brief
Tell us which layer hurts.
One email starts it. We ask about the layer and the org after. The assessment is free: a short read of the org and a written note. If you go ahead, the first build week is a fixed AED/SAR 14,900 (about USD 4,050), credited in full when you sign the build within 60 days.