Overview & Core Concepts
Welcome to the AETHER platform documentation. AETHER lets you build, test, publish and integrate AI agents — backed by your own knowledge, connected to real tools, and governed centrally — without managing the underlying orchestration yourself.
This section is written for developers. It walks through the concepts you'll meet across the platform, then the rest of the guide takes each one in depth with worked examples and screenshots.
What is an agent?
An agent is a configured AI assistant. Each agent bundles together:
- a persona — its name, tone, response length, creativity and language;
- system instructions — the prompt that defines its job and boundaries;
- knowledge — documents, web pages and files it can retrieve from (RAG);
- tools — external systems it can call via the Model Context Protocol (MCP);
- appearance — avatar, colours and branding for published surfaces.
You build an agent once, then test it, publish it, and expose it over a chat widget, the REST API, or Microsoft Teams.
Core concepts
| Concept | What it means |
|---|---|
| Agent | A configured AI assistant you build and publish. |
| Conversation | A stateful chat session with an agent; every message, tool call, citation and token is logged. |
| Knowledge source (RAG) | A PDF, text file or URL that is chunked, embedded and indexed so the agent can retrieve relevant passages. |
| Tool (MCP) | An external capability — an API, database or function — the agent can call mid-conversation. |
| Workflow | A visual, branching script that drives a conversation through defined steps. |
| Environment | A stage in an agent's lifecycle: Development → QA → Release. |
| Invocation | A single model call, recorded with its token usage and cost. |
How the pieces fit together
A typical request flows like this:
- A user sends a message into a conversation.
- The agent retrieves relevant passages from its knowledge sources (RAG).
- If the task needs live data or an action, the agent calls a tool (MCP).
- The model produces a response; the invocation (tokens + cost) is recorded.
- The full exchange — text, tool calls and citations — is stored for auditing.
The platform at a glance
- Dashboard — your home base: every agent you own or can administer, plus quick stats.
- Agent builder — configure persona, instructions, knowledge, tools and appearance.
- Workflow designer — drag-and-drop branching conversations, with RAG and MCP in any step.
- Test console — chat with an agent and inspect invocation logs, tokens and cost.
- Publishing — promote an agent through environments and expose it (widget, API, Teams).
- Admin portal — govern organisations, deployments, budgets, models and usage. See the Administration docs.
Where to go next
- New here? Start with Signing In & the Dashboard, then Create Your First Agent.
- Integrating from code? Jump to the API Reference.
- Administering the platform? See Administration → Overview & Admin Levels.