Fundamentals¶
Everything you need to build, deploy and run an agent: setup, the build lifecycle, the core components agents use, and the reference architectures.
Install & Setup¶
-
Prepare an environment for building and running agents: prerequisites, install scripts, their order and service configuration.
Build¶
-
Agent Profiles: Specification & Customisation
The schema that describes an agent: identity, persona, prompting rules, execution policy, integrations, resources and governance.
-
What an agent is made of, and how to write one.
-
Store an agent's specification in the subjects registry so it can be discovered and deployed.
-
Turn a registered agent into running workloads on Kubernetes.
-
Run and load-balance many instances of an agent, and scale them with demand.
-
How agents call LLMs and other AI models.
-
Test an agent in three layers, from fastest to most realistic.
-
Let a deployed agent take work from a shared task exchange.
Core Components¶
-
How agents store and recall memory through the agents SDK.
-
Agentic Function & Tool Calling
How agents discover and call functions and tools.
-
Break work into repeatable, inspectable steps: static and dynamic workflows, and calling them from agents.
-
Generate code with an LLM and, optionally, execute it at runtime.
-
One interface for generating embeddings from different backends.
-
Chat, peer-to-peer and agent-to-agent (A2A) communication between agents.
-
Agentic Task Exchange & Bidding
How agents post, take and bid on work through task exchanges.
-
Graph-based knowledge storage for agents.
-
Vector storage and similarity search for agents, with Milvus and pgvector backends.
-
Long-term storage for reports, files, model outputs, checkpoints and other artefacts.
-
A file-system interface over agent memory.
Architectures¶
-
Architecture diagrams for the agent platform.