Architecture
The platform
Asgard AI is an enterprise SaaS platform for building LLM applications, aimed at generative AI. Three products sit at its centre:
| Product | Where it sits | What it does |
|---|---|---|
| Odin Studio | The AI enablement console | Visual workflow design, AI resource management, version control |
| Mimir Data Insight | AI data insight | Generative analysis, intelligent queries, decisions from data |
| Sindri Agent Hub | The agent execution engine | Conversation, AI agent dispatch, fast deployment |
The architecture
Core concepts
Project and workflow
- Project: the top-level unit, holding one or more workflows
- Workflow: how an AI application runs, built by connecting processors
- Processor: the smallest unit of execution inside a workflow
Channel and session
- Channel: the conversation between a user and the AI, identified by
customChannelId - Session: the memory of a conversation within one channel, kept until it is reset by hand
Knowledge base (RAG)
- Accepts manually uploaded documents, a web crawler and a database loader
- Builds a vector index automatically, for the Retrieve Knowledge processor to query
- Supports segment settings and tag management
How a request flows
A typical AI conversation request goes through the following.
The processors
| Category | Processor | What it does |
|---|---|---|
| Flow | Entry | Where the workflow starts |
| Exit | Where the workflow ends | |
| Router | Conditional branching | |
| Message | Push Message | Sends a message |
| Listen Message | Receives user input | |
| Model | LLM Completion | Calls a large language model |
| Stream LLM Completion | A streamed AI response | |
| Generate Embedding | Turns text into vectors | |
| Query | SQL | Queries a database |
| Retrieve Knowledge | Searches a knowledge base | |
| Action | Update Context | Updates the value of a variable |
| Execute Script | Runs JavaScript | |
| API | HTTP Request | Calls an external API |
| AutoTool | Validate Payload | Validates the input |
| Response | Formats what is returned |
Ways to integrate
Asgard supports several, covered in the integration documentation:
| Method | What it is | Where it fits |
|---|---|---|
| REST API | Calling the SSE API directly | Back-end integration, a custom front end |
| Asgard SDK | The JavaScript / React SDK | Getting a web application connected quickly |
| iFrame | An embedded chat window | Adding chat to an existing site quickly |
| Bot Platform | LINE / Slack / Discord / Telegram | Messaging platforms |
| MCP Server | Model Context Protocol | Extending an AI agent's tools |
Next
- Quick start — build your first AI workflow from nothing
- API reference — the full RESTful API
- Processor reference — the specification of every processor