Instructions, Assets, and Sub Agents
When Aissistant responds to a user’s inquiry, it doesn’t just guess the answer — it intelligently pulls context and instructions from three main sources: workspace instructions, assets, and sub-agents. Each plays a unique role in shaping how Aissistant understands and replies. Let’s break down the difference.
🔧 1. Workspace Instructions – The Global Rulebook
What they are: These are the global instructions that guide Aissistant’s behavior across all conversations.
How they work: Workspace instructions are automatically included every time Aissistant generates a response. Because they apply globally, they are meant to be brief, high-level, and always relevant.
Example:
“Always answer in a friendly and professional tone.” “Use markdown for formatting.”
📚 2. Assets – The Business Knowledge Base
What they are: Assets are the documents, web pages, or resources that contain your business-specific knowledge.
How they work: Assets are stored in our system as trunks, which are smaller segments of the original content. For instance, if an asset is a large manual, it will be split into multiple trunks. When a user asks a question, instead of returning the whole asset, Aissistant searches through these trunks to find relevant information. If a useful trunk is found, it’s added to the response context. You can think of this as Aissistant’s built-in search engine, retrieving just the right slice of knowledge to answer the query.
Example:
An FAQ page, product documentation, or internal troubleshooting guide.
🤖 3. Sub Agents – The Specialized Experts
What they are: Sub-agents are scenario-specific configurations that give Aissistant special instructions for certain types of requests.
How they work: If a user’s message matches a scenario defined in a sub-agent (like billing issues, shipping delays, etc.), the sub-agent's instructions are added to guide the reply. If the sub-agent has assets linked to it, Aissistant will also search trunks in those assets and include any relevant trunks — improving accuracy and reducing latency.
Example:
A sub-agent called
shippingmight handle shipping questions and only search shipping policy documents.
🧩 Putting It All Together
When Aissistant replies, it checks:
Workspace Instructions – Always included
Sub-Agent – If triggered, instructions are included, linked asset searched and included if relevant
Asset – Searched and included if relevant
By combining these layers, Aissistant delivers fast, relevant, and accurate responses tailored to each situation — just like a well-trained human support agent.
Last updated