How to Maximize AVCodex Capabilities
Get the most out of your AVCodex agents by mastering the features that matter most for AV teams.
Your AVCodex agent is only as good as the knowledge behind it. Understanding how AI processes different types of knowledge is the foundation for building effective agents (whether that is a programmer help bot, a service desk agent, or a pre-sales RFP assistant).
There are three types of knowledge your agent can use:
- Base model knowledge: what the AI already knows from its training data.
- Knowledge sources: documents, files, and URLs you upload (manufacturer manuals, AS-builts, internal SOPs, programming guides).
- Prompt instructions: direct guidance you write in your system prompt.
Each one serves a different purpose. Master all three and your agent gets dramatically more useful.
The three types of knowledge#
The AI's base knowledge is broad but generic. It knows about the world. It does not know about *your* firm, *your* clients, *your* truck stock, *your* programming standards, or *your* specific install patterns.
That is where knowledge sources come in.
Why upload knowledge on top of the AI's existing knowledge#
Uploading your own knowledge sources gives your agent:
- Accuracy: answers grounded in your actual content, not AI guesses.
- Specificity: details about your gear, your manufacturer mix, your processes.
- Currency: up-to-date information the base model does not have. New firmware, new manufacturer guides, new internal standards.
- Authority: responses that reflect your firm's voice and engineering opinion.
The AI's training data has a cutoff date. Your knowledge sources do not. Upload your latest commissioning checklist and your agent stays current.
Knowledge source vs prompt: the technical difference#
Knowledge sources are retrieved on demand. When a user asks a question, AVCodex searches your uploaded documents and pulls relevant chunks into context. This is RAG (Retrieval-Augmented Generation).
Prompts are always present. Your system prompt is included in every conversation, regardless of what the user asks.
| Feature | Knowledge sources | Prompts |
|---|---|---|
| When used | Retrieved when relevant | Always included |
| Best for | Large amounts of reference content (full manufacturer manuals, every AS-built) | Personality, rules, behavior |
| Size limit | Large (many documents) | Small (a few thousand words) |
| Updates | Upload new files anytime | Edit in Build settings |
Rule of thumb. Put facts and reference material in knowledge sources. Put personality, tone, and behavioral rules in your prompt.
Use scheduled refetching to keep documents and crawled URLs current without thinking about it. A useful pattern: refetch your manufacturer help-site URLs weekly and your internal SOP repository on every commit. Watch the in-platform tutorial for the click path.
See the section above on Knowledge source vs prompt for the technical breakdown. The short version: knowledge sources scale, prompts steer behavior.
Mastering the basics of Markdown for both instructions and knowledge sources can make a real difference in accuracy and ease of editing.
What is Markdown#
Markdown is a simple way to add formatting (bold, italics, lists, headings) to plain text. It is widely used in documentation and AI tools to make instructions clearer and more organized.
Why use Markdown in AI instructions#
- Clarity: well-formatted instructions are easier for both humans and AI to follow.
- Emphasis: you can highlight important steps or warnings (think safety language in commissioning instructions).
- Organization: headings and lists break information into manageable sections.
Basic Markdown formatting#
A few easy ways to use Markdown:
Headings. Use # to tell your agent which sections belong where in the hierarchy. Heading 2 (##) sits under Heading 1 (#). Heading 3 (###) sits under both.
# Main Section Title
## Section Title
### Subsection
#### Sub-subsectionBold and italic.
- Bold for important words.
- *Italic* for emphasis.
Lists.
- Step 1.
- Step 2.
- Step 3.
Links. If you want more examples of Markdown and HTML, the Markdown Guide is a solid reference.
How to use Markdown in your AI instructions#
When you write instructions for your agent, use Markdown to:
- Build step-by-step guides with clear sections and hierarchy (commissioning runbooks, troubleshooting trees).
- Highlight key actions or warnings in bold.
- Organize long instructions with headings.
Example, plain instruction:
Please collect tech feedback and send a confirmation message.
Example, with Markdown:
## Collect Feedback
- Ask the tech for their feedback on the dispatch.
- **Confirm receipt** with the tech.
- Save the feedback for future service improvements.Markdown helps your agent communicate more clearly. No code required.
You can share images and GIFs as part of your agent's flow (visual install steps, photos of correctly terminated cable, animated rack walkthroughs). Use this when a picture is faster than a paragraph.
*Video coming soon.*
*Content coming soon.*
*Content coming soon.*
You can use the image generation tool to create variations of the same image for different platforms or use cases (a wide rack photo for a case study page, a vertical version for social, a square one for an email header).
You can structure an agent to take in information from a junior programmer (their experience level, the systems they want to learn) and produce a personalized lesson plan. Useful for new-hire onboarding at integrators and for cross-training existing programmers from Crestron SIMPL onto Q-SYS Lua.
You can create video using Gemini 2.5 inside AVCodex. Useful for room walkthrough videos, manufacturer comparison clips, or quick visual install guides.
*AVCodex · Your AV expertise. Amplified by AI.*