The Agentic Framework
The layer that acts on the brain: agents, skills, prompts, chat sessions, and multi-agent coordination.
The agentic framework is everything that reads from and writes to the brain. The brain holds the work (tasks, memories, and files); the framework is the set of moving parts that does the work against it. This section covers each part and how they fit together.
The building blocks
An agent is a single unit: one versioned prompt paired with one model. The model can be overridden per call, or the agent can inherit the platform default so operators change one setting instead of many. Skills and prompts are the reusable pieces an agent draws on. Chat sessions are the persistent, resumable conversations where agents and people meet.
Coordination
Individual agents rarely work alone. An assistant is several agents working together, so the framework includes how they coordinate: parallel multi-agent spawning over an agent-to-agent protocol, and blog posts used as handoffs between agents and humans on a task. Agents also work within the limits of the model context window, so the framework covers how work is scoped and passed along.
Explore the parts
Agents
A versioned prompt plus a model choice, with per-call overrides and default inheritance.
Skills
Reusable capabilities an agent can call on to do focused work.
Prompts
Versioned instructions that define how an agent behaves.
Chat
Persistent, nameable, resumable sessions that can link to tasks.
Coordination
Multi-agent spawning, agent-to-agent handoffs, and working within the context window.