# The Alchemist Copilot

Your AI builder assistant. It analyzes conversations, configures agents, generates assets, and runs your AVCodex workspace through plain English.

> **Note:** "Alchemist" names two related things. This page covers the **Alchemist copilot**, the chat assistant inside the AVCodex agent builder. For the ticket-driven autonomous development platform, see [Alchemist Tickets](/docs/guides/alchemist-tickets).

---

The Alchemist is the AI builder assistant baked into AVCodex. It can read your agent's chat logs, change agent configuration, draft content, manage knowledge sources, and more, all through natural language. Think of it as a co-programmer who knows every screen in the platform and can drive them on your behalf.

## Where to find it

The Alchemist panel appears in two places:

- **Dashboard**: The panel on the main dashboard runs in workspace mode. It can read analytics across every agent in your AVCodex account, compare performance, and jump you into any agent's builder.
- **Agent Builder**: When you are editing a specific agent, the Alchemist panel has full access to that agent's configuration. It can read and change the system prompt, knowledge sources, integrations, tags, and the rest.

## What it can do

### Conversation analysis

The most useful job: ask the Alchemist to read your chat logs and tell you how to improve the agent. It will:

1. Sample recent conversations across categories (short and abandoned, long and engaged, recent, random).
2. Spin up a dedicated analysis agent with a fresh context window focused on the conversation data.
3. Identify frustration patterns, knowledge gaps, error patterns, and what is working well.
4. Return prioritized, specific recommendations grounded in real chat examples.

**Try asking:**

- "Analyze my conversations and tell me how to improve this agent."
- "What are integrators struggling with when they ask about Q-SYS Lua?"
- "Look at the last 7 days of chats and find knowledge gaps."

After the analysis, the Alchemist can act on its own recommendations: rewrite the system prompt, add knowledge sources, or set up new tags.

### Analytics and insights

Get answers about agent performance without clicking through dashboards:

| What you can ask | What it does |
|---|---|
| "How many conversations did I have this week?" | Returns session counts, message volume, and trends. |
| "What are field techs asking about?" | Samples recent first messages and clusters them by topic. |
| "Show me my top consumers." | Consumer report with engagement, leads, and email capture. |
| "Which agent is getting the most traffic?" | Per-agent overview with chat counts and consumer counts. |
| "Search for conversations about Crestron error codes." | Full-text search across all chat messages. |
| "What tags are firing most often?" | Tag distribution with percentages and sample messages. |

All analytics tools accept a date range ("last 7 days", "last 90 days") and can be scoped to a specific agent.

### Agent configuration

The Alchemist can read and modify every part of an agent's configuration:

- **System prompt**: "Update the system prompt so the agent answers in the voice of a senior commissioning engineer and routes warranty questions to the parts team."
- **Model selection**: "Switch this agent to Claude Sonnet."
- **Brand and styling**: "Set the primary color to our integrator brand teal and generate a logo."
- **Knowledge sources**: "Add this URL as a knowledge source," or "Add these five Biamp datasheet PDFs to the knowledge base."
- **Custom Actions**: "Create an HTTP action that hits our ticketing API to look up an open dispatch."
- **Tags**: "Set up tags for warranty claims, RMA requests, and project handoff questions."
- **Auth settings**: "Require users to log in before chatting."
- **Voice settings**: "Turn on voice mode with a clear, neutral voice."
- **Heartbeat**: "Send a weekly check-in to commissioning engineers asking how their current cutover is going."
- **Publishing**: "Publish this agent live."

### Content generation

Create visual assets and web content through conversation:

- **Logos**: "Generate a logo for a Crestron-focused programming help desk." Creates an AI logo and applies it.
- **Custom CSS themes**: "Design a dark control-room theme with amber accents and squared corners." Generates and applies the theme through Style Studio.
- **Landing pages**: "Build a landing page that explains what this agent does for our integrator clients." Creates a multi-page site through AVCodex Sites.
- **Custom components**: "Create a rack elevation component with three populated bays." Generates an interactive HTML, CSS, and JS component.

### Integration management

Build and deploy integrations without code:

- **MCP servers**: "Create an MCP server that connects to our project management API so the agent can pull active room counts on a job."
- **Deploy and test**: "Deploy the integration and test the list_projects tool."
- **OAuth setup**: Walks you into OAuth configuration flows when an integration needs them.

### Marketplace and Templates

Find and use pre-built blueprints and composable Skills:

- **Browse**: "Show me Templates for service desk agents."
- **Create from Template**: "Create a new agent from the Field Tech Support Template."
- **Apply Skills**: "Add the Empathetic Listener Skill to this agent." Skills are composable prompt modules that layer onto your agent.

### Advanced reasoning

For complex questions, the Alchemist can think harder:

- **Deep thinking**: Breaks a problem into structured steps. "Plan how I should restructure this agent to handle both pre-sales spec questions and post-install support."
- **Deep research**: Spawns parallel web research agents to pull authoritative information. "Research best practices for AI agents that triage AV control system error codes."

## Example workflows

### "Improve my agent" workflow

The most common, most useful loop:

**1. Ask for analysis**

"Analyze my conversations and tell me what I should improve."

**2. Review the findings**

The Alchemist returns categorized issues: frustration patterns, knowledge gaps, error patterns, and what is working well. Each item includes real conversation examples and a specific fix.

**3. Implement the fixes**

"Go ahead and add the AS-built documentation for the Acme HQ project to the knowledge base," or "Update the system prompt to handle warranty escalations." The Alchemist uses its builder tools to make the changes directly.

**4. Publish**

"Publish the agent." Pushes the improvements live.

### New agent setup workflow

**1. Create from a Template or from scratch**

"Create a service desk agent for a managed AV firm," or "Use the Field Tech Support Template."

**2. Configure**

"Set the system prompt to help our field techs troubleshoot Crestron, Q-SYS, and Extron systems. The company is Acme Integration and we run service for 80 enterprise clients."

**3. Add knowledge**

"Add these URLs as knowledge sources: support.crestron.com error-codes, qsys.com/help, our internal commissioning checklist PDF."

**4. Brand it**

"Generate a logo and set the brand color to #0E5E63."

**5. Publish**

"Publish the agent."

### Cross-agent analytics workflow

From the dashboard, you can analyze performance across all your agents:

1. "Which of my agents is getting the most conversations?"
2. "Show me the chat analytics for the last 30 days across all agents."
3. "What topics come up most often across the field tech agent and the spec engineer agent?"
4. "Open the Field Tech Support agent so I can make changes." Drops you into the agent builder with the Alchemist panel open.

## Dashboard mode vs agent mode

| Capability | Dashboard | Agent Builder |
|---|---|---|
| Analytics (all agents) | Yes | Yes (scoped to current agent) |
| Conversation analysis | Yes (any agent) | Yes (current agent) |
| Browse Templates | Yes | Yes |
| Create agent from Template | Yes | Yes |
| Read agent config | After selecting an agent | Automatic |
| Modify agent settings | After selecting an agent | Automatic |
| Generate logos, CSS, sites | After selecting an agent | Automatic |
| DOM interaction (uploads, OAuth) | After navigating to agent | Yes |

In dashboard mode, the Alchemist calls **select_app** when it needs to work with a specific agent. You can also tell it directly: "Select my Field Tech Support agent and update its system prompt."

## Tips

- **Be specific.** "Add a knowledge source about our RMA policy" beats "improve my knowledge base."
- **Chain requests.** The Alchemist holds context, so you can say "Now also add the Q-SYS commissioning guide" right after adding a Crestron one.
- **Run analysis first.** Before making changes, ask the Alchemist to analyze conversations. Data-driven fixes beat guesswork.
- **Filter by date.** "Analyze conversations from the last 7 days" surfaces recent patterns. "Last 90 days" gives the wider picture.
- **Focus the analysis.** "Analyze conversations with a focus on commissioning issues" tells the analysis agent where to look.

> **Note:** The Alchemist is available on all plans. Some tools it uses (custom CSS, voice cloning, or advanced integrations) may require specific plan tiers.

*AVCodex · Your AV expertise. Amplified by AI.*