Core Concepts
Karyam is built around a small set of foundational concepts.
Understanding these concepts makes it easier to design, operate, and scale AI systems inside your organization.
The Karyam Mental Model
Section titled “The Karyam Mental Model”Workspace ↓Models ↓Agents ↓Skills + MCP Servers ↓AI Flows ↓Approvals ↓RunsAlongside execution infrastructure:
Files ↓Listeners ↓Embeddings ↓Vector Databases ↓RAG RunsAnd operational capabilities:
TeamsRBACAudit LogsAPI TokensDashboardWorkspace
Section titled “Workspace”Everything in Karyam lives inside a Workspace.
A Workspace contains:
- Agents
- AI Flows
- Models
- Skills
- MCP Servers
- Files
- Vector Databases
- Teams
Think of a Workspace as an isolated AI environment.
Models
Section titled “Models”Models provide reasoning and generation capabilities.
Examples include:
- GPT
- Claude
- Gemini
- Bedrock
- Ollama
Models power:
- Agents
- AI Flows
- Embeddings
Agents
Section titled “Agents”Agents are intelligent runtime entities capable of:
- Understanding requests
- Using tools
- Accessing knowledge
- Collaborating with users
Agents are responsible for decision making.
Skills
Section titled “Skills”Skills allow AI systems to interact with external systems.
Examples include:
- PostgreSQL
- SSH
- Redis
- GitHub
Skills provide capabilities.
MCP Servers
Section titled “MCP Servers”MCP Servers provide standardized access to external tools and systems.
They extend AI capabilities without requiring custom integrations.
AI Flows
Section titled “AI Flows”AI Flows orchestrate business processes.
They provide:
- Control flow
- Approvals
- Automation
- Governance
Flows coordinate execution.
Files provide enterprise context.
Uploaded documents become retrievable knowledge through the RAG pipeline.
Listeners
Section titled “Listeners”Listeners automate knowledge ingestion.
When new files arrive, listeners trigger:
- Chunking
- Embeddings
- Vector storage
Vector Databases
Section titled “Vector Databases”Vector Databases store embeddings and power semantic search.
Examples include:
- Qdrant
- pgvector
RAG Runs
Section titled “RAG Runs”RAG Runs provide visibility into ingestion and retrieval operations.
Runs represent actual executions of Agents and AI Flows.
Every interaction becomes a Run.
Runs provide:
- Logs
- Token usage
- Cost analytics
- Debugging information
Governance
Section titled “Governance”Karyam includes enterprise governance capabilities:
- Teams
- RBAC
- Audit Logs
- Approvals
These capabilities enable safe AI adoption.
The Final Picture
Section titled “The Final Picture”Knowledge +Agents +Skills +Workflows +Governance +Observability ↓Production AI SystemsThis is the Karyam approach.
Build AI Systems. Not Hacks.