Console Guide

The Agent Orchestrator dashboard at console.appxen.ai provides a visual interface for creating, running, and monitoring workflows.

Overview Page

The overview shows your recent executions and workflow count at a glance. Use the sidebar navigation to access Workflows, Executions, and the Gallery.

Workflow Copilot

The AI copilot is built into the workflow editor. Toggle it with the Agent button in the toolbar — it opens by default when you create or edit a workflow.

Describe what you want in plain English and the copilot generates the complete workflow markdown, including the right Tools: directives based on your connected MCP servers. The generated markdown is applied directly to the editor below, where you can review and refine it.

How to Use

• Describe your workflow — Type a prompt like “Create a workflow that researches a topic and writes a report”
• Iterate — Chat back to refine: “Add an approval gate before the final step” or “Switch the researcher to Haiku”
• Switch models — Toggle between Haiku (faster, lower cost) and Sonnet (more capable) for the copilot itself using the model selector below the input
• Edit alongside — The editor stays visible below the chat. Toggle Edit on to make manual changes, or off to preview the rendered markdown
• Resize — Drag the handle between the chat and editor to adjust the split

The copilot is tool-aware — it knows which MCP servers are connected to your account and uses their aliases in Tools: directives automatically. This prevents common errors like referencing a server that isn’t configured.

Workflow Editor

Navigate to Workflows and click New Workflow (or open an existing one) to launch the editor.

CodeMirror Editor

The built-in editor provides:

• Syntax highlighting — Markdown headings, code blocks, and inline formatting are color-coded
• Autocomplete — Trigger with Ctrl+Space for DSL sections (## Inputs, ## Agents, etc.), model names, and variable placeholders
• Live validation — Checks for required sections and common mistakes as you type
• Dark theme — Matches the AppXen design system

Compiled Plan Preview

After saving, the compiled execution plan appears below the editor. This shows the resolved agents, steps, dependencies, and any compiler notes. Review the plan to verify the compiler understood your intent before running.

Template Gallery

The Gallery page offers pre-built workflow templates across five categories. Search by name or filter by category:

Decision Making

• Devil's Advocate — Agents argue both sides of a decision, then a judge synthesizes a balanced verdict

Content Creation

• Blog Post Pipeline — Research, draft, and edit to publication quality
• RFC Writer — Generate a full RFC document with problem analysis, proposed solutions, and trade-offs
• SEO Landing Page Generator — Analyze a product page's design, then generate a matching SEO-optimized HTML landing page

Research

• Explain Like I'm... — Get explanations at three levels, then a master teacher synthesizes the best one
• Meeting Prep Brief — Research attendees, company, and agenda to prepare a comprehensive brief
• Property Appraisal Report — Research a property via web search, gather comparables and market data, then generate a professional appraisal report (uses Tavily)

Analysis

• SWOT Analysis — Parallel analysis of strengths, weaknesses, opportunities, and threats
• Code Review Panel — Multiple specialist reviewers (security, performance, architecture) with a lead synthesizer

Developer Tools

• CloudWatch Log Analysis — Analyze CloudWatch logs to identify errors and patterns (uses MCP tools)
• Neon Database Seed — Generate realistic seed data for a Neon database (uses MCP tools)
• Log-to-Fix Pipeline — Detect errors in logs, find root causes, create a fix PR, and write an incident report (uses MCP tools)

Click Use Template to create a new workflow from a template. You can then customize it in the editor.

Giving Agents Web Access

By default, workflow agents use LLM knowledge only — they cannot access the web. To give agents the ability to search the internet for current information, connect a web search server through the MCP Gallery.

Setting Up Tavily Web Search

1. Get a Tavily API key — Sign up at tavily.com (free tier includes 1,000 searches/month)
2. Add the server — In the dashboard, go to MCP Gallery and add the Tavily server. Enter your API key when prompted.
3. Use it in workflows — Add Tools: tavily to any agent that needs web access

### Researcher
Expert at finding and synthesizing current information from the web.
Model: Sonnet
Tools: tavily

Tavily is optimized for AI agents — it returns concise, AI-ready text summaries rather than raw HTML, making it ideal for workflow agents that need to reason over search results. Agents can combine web search with other tools:

### Researcher
Searches the web for current data and cross-references with internal docs.
Model: Sonnet
Tools: tavily, rag-engine

Running Workflows

Click Run on any workflow page. If the workflow has inputs, a dialog appears to fill them in. After confirming, you're redirected to the execution detail page.

Execution Detail

The execution page provides real-time monitoring:

• Progress bar — A segmented bar showing which step is currently executing, with step names labeled below.
• Step list — Each step shows its status (pending, running, completed, failed). Click to expand and see the step's output, token usage, and model used.
• Token summary — Total input/output tokens and step count.
• Stop button — Cancel a running execution at any time.

Approval Gates

When a workflow hits an approval gate, the step shows Approve and Reject buttons. The execution pauses until you make a decision. Rejecting a step stops the workflow.

Executions List

The Executions page shows all your workflow runs with workflow name, status, start time, duration, and step progress. Click any row to see the full execution detail. The page auto-refreshes while running executions are in progress.

• Search — Filter executions by workflow name
• Status filter — Filter by status (All, Running, Completed, Failed, Stopped, Waiting)
• Source filter — Filter by execution source (All, Manual, Scheduled)

Scheduled executions are marked with a clock icon. Each execution card includes a three-dot menu for quick actions like re-running the workflow or deleting the execution.

Output & Download

When an execution completes, the final output renders at the bottom of the detail page. The steps section is collapsible — collapse it to give the output the full viewport height. Content is auto-detected and rendered appropriately:

• HTML — Live preview in a sandboxed iframe with a source/preview toggle
• SVG — Rendered inline with a source/preview toggle
• Markdown — Formatted prose with headings, lists, tables, and code blocks
• JSON — Pretty-printed with syntax highlighting
• Code — Monospace code block with language hint
• Plain text — Auto-detected as markdown when formatting is present, otherwise displayed as-is

Output Actions

The output toolbar provides several actions:

• Download — Save the output as a file. The extension matches the content type (.html, .md, .json, etc.)
• Share — Generate a public link to the output hosted at go.appxen.ai. Anyone with the link can view the output. You can revoke access at any time.
• Refine — Give the AI a follow-up prompt to improve the output. The original context and all step results are preserved. Each refinement creates a new version.
• Save to Knowledge Base — Save the output to your RAG Engine knowledge base for future retrieval by other workflows or search.
• Copy — Copy the raw output to your clipboard.

Version History

Each refinement creates a new output version. Click the version dropdown next to the output heading to browse previous versions. Each entry shows the refinement prompt and timestamp. Select any version to view its output, or return to the latest version.

Editor Shortcuts