AI-Assisted Development: Natural Language to Business Workflows

The promise of "describe what you want and get a working app" is easy to overstate. AuraBoot takes a narrower path: use natural language where it helps with querying, operating, and guiding business applications, while keeping the actual data model, commands, permissions, and audit path explicit.

The AI Stack

AuraBoot's AI capabilities are layered, not monolithic:

Layer 1: AuraBot — The Conversational Copilot

AuraBot is an AI assistant embedded in the AuraBoot interface. It understands your data model, permissions, and business context. Ask it a question, and it queries your actual data:

You: "How many open leads do we have from LinkedIn this month?"

AuraBot: "You have 23 open leads from LinkedIn created in March 2026. The breakdown by status: 12 New, 8 Contacted, 3 Qualified. Would you like me to show the full list?"

This isn't a generic chatbot. AuraBot knows your models, fields, dictionaries, and relationships because they're all defined in the DSL. It translates natural language into API calls against your real data.

Layer 2: Agent Control Plane

The Agent Control Plane (ACP) manages AI agents that can perform multi-step operations autonomously. Unlike simple chatbots, agents have:

• Skills: Registered capabilities (query data, create records, send notifications)
• Tools: Bound platform functions the agent can invoke
• Memory: Conversation history and context that persists across sessions
• Guardrails: Permission checks and entitlement limits on what agents can do

An agent might handle: "Create a follow-up task for every lead that hasn't been contacted in 7 days, assign it to the lead owner, and set the due date to tomorrow."

This single instruction triggers the agent to:

1. Query leads with status = NEW and created_at < 7 days ago
2. For each lead, create a task record via the command pipeline
3. Set the assignee from the lead's owner field
4. Set the due date to tomorrow
5. Report the results

All operations respect the same permission model, audit trail, and validation rules as manual actions.

Layer 3: Agent-Readable Knowledge Documents

The knowledge document layer lets you manage company documents — PDFs, Word files, spreadsheets, markdown — and make governed document content available to AI through platform read tools:

1. Upload: Documents are created or imported into the knowledge module
2. Govern: Versioning, status, access level, and project links remain part of the command pipeline
3. Read: Agents access permitted document content through platform tools
4. Answer: AuraBot grounds responses in documents it is allowed to read

This means AuraBot can answer questions like "What's our return policy for enterprise customers?" by referencing your actual policy document, not hallucinating an answer.

Multi-Provider Support

AuraBoot doesn't lock you into a single AI provider. The platform supports 8 LLM providers:

Provider	Models
OpenAI	GPT-4o, GPT-4o-mini
Anthropic	Claude 3.5 Sonnet, Claude 3 Opus
Google	Gemini Pro, Gemini Flash
DeepSeek	DeepSeek-V3, DeepSeek-R1
Zhipu AI	GLM-4
MiniMax	abab6.5
Moonshot	Moonshot-v1
Ollama	Any local model

You configure providers at the platform level and assign them to agents. Different agents can use different models — a coding assistant might use Claude, while a data analyst uses GPT-4o. Switching providers requires changing a configuration, not rewriting code.

Practical AI Workflows

Data Analysis

You: "Show me our top 10 customers by revenue this quarter, compared to last quarter."

AuraBot generates the appropriate Named Query, fetches the data, and presents a comparison table with growth percentages. If you want a chart, say "make that a bar chart" and it renders one.

Record Creation

You: "Create a new product called 'Enterprise License' priced at $999/month in the Software category."

AuraBot invokes the product__create command with the appropriate fields, passing through the full 20+ stage pipeline including validation, permission checks, and audit logging.

Bulk Operations

You: "Mark all overdue invoices from last month as 'Escalated' and notify the account managers."

AuraBot identifies the affected records, confirms the count with you, then executes bulk updates followed by notifications — all through the platform's standard command and notification systems.

Report Generation

You: "Generate a monthly sales report for March with revenue by product category and a pipeline forecast."

AuraBot queries the relevant data sources, builds a report using the Report Designer's DSL, and provides a downloadable PDF.

How DSL Enables AI

The practical reason this works is that AuraBoot's DSL gives AI features a structured map of the application. Here's why:

Structured Metadata

Every model, field, command, and relationship is described in JSON. AI agents don't need to reverse-engineer database schemas or read source code. They have a complete, structured description of the entire application.

Bounded Operations

Commands define exactly what operations are possible and what parameters they accept. An AI agent can't accidentally execute an undefined action — it can only invoke registered commands that pass through the full validation pipeline.

Permission Enforcement

AI agents operate within the same RBAC system as human users. An agent running on behalf of a sales rep can't access finance data, and a free-tier tenant's agent can't invoke enterprise features.

Natural Language Mapping

Field names, model descriptions, and dictionary labels provide the vocabulary for natural language understanding. When a user says "leads," the system maps it to crm_lead. When they say "qualified," it maps to the QUALIFIED dictionary value.

Setting Up AI

Configure a Provider

Add your API key in the Agent Control Plane settings:

1. Navigate to Mission Control > Providers
2. Select a provider (e.g., OpenAI)
3. Enter your API key
4. Test the connection

Enable AuraBot

AuraBot is available by default. Click the chat icon in the bottom-right corner of any page. It automatically has access to all models and commands your user role permits.

Create a Custom Agent

For specialized workflows, create a custom agent:

1. Navigate to Mission Control > Agents
2. Define the agent's skills and tool bindings
3. Configure the LLM provider and model
4. Set permission boundaries
5. Deploy

Custom agents can be exposed as API endpoints, embedded in pages, or triggered by events.

What AI Can't Do (Yet)

Transparency matters. Here's what AuraBoot's AI doesn't handle today:

• Schema design: AI can query and manipulate data, but designing the data model is still a human task
• Complex business logic: Multi-step workflows with conditional branching still require DSL configuration
• Visual design: Page layouts and dashboards are configured through the designer, not natural language
• Training on your data: AuraBot uses retrieval (RAG), not fine-tuning. Your data never leaves your infrastructure for training

These boundaries are by design. AI handles what it does well (language understanding, data querying, routine operations) while humans retain control over architecture and business rules.

Conclusion

AuraBoot's AI integration is wired into the same DSL, command, permission, and audit paths as the rest of the platform. The goal is practical assistance — data questions, workflow help, knowledge lookup, and controlled operations — without forcing one LLM vendor.

The result is modest but useful: less context hunting, fewer repetitive clicks, and a clearer path from a natural-language request to an audited business operation.

Try AuraBot in your own AuraBoot instance and tell us where the beta still feels rough.