Demo
At MangoBytes, we set out to build something ambitious: an AI platform that could reason, collaborate, and take meaningful actions across complex oil & gas operations.
We imagined a system where intelligent agents could analyze production trends, explain cost overruns, forecast decline curves, and surface insights across leases, wells, and fields—all while maintaining transparency, control, and security. That vision led to PIXIE.
Today, we’re excited to announce PIXIE—our new multi-agent platform designed to bring intelligent automation to real-world challenges in energy and oil & gas operations.
What is PIXIE?
PIXIE is a framework where multiple AI agents—each specialized in oil & gas domains—can collaborate, access real-time production and financial data, and take action across tools and systems.
Unlike a single chatbot interface, PIXIE enables multi-agent collaboration, secure execution, and traceable decisions—making it ideal for tasks like production forecasting, revenue analysis, LOE tracking, and regulatory compliance.
We built PIXIE for how field teams, engineers, analysts, and executives actually work—where delivering insights means pulling from accounting systems, field reports, spreadsheets, and public datasets, then delivering results via Teams, email, or dashboards.
Why We Built This
Here’s what we saw in the field:
- Production engineers don’t just want “results.” They want reasoning steps, assumptions, context, and visualizations.
- Finance teams need forecasts—but also want to tweak inputs, compare price decks, and export to Excel.
- Operations leaders want to audit decisions, enforce access controls, and explain what agents are doing.
So we engineered PIXIE to handle:
- Multi-step reasoning (Chain of Thoughts)
- Agent collaboration across oil & gas domains
- Real-time data access with scoped permissions
- Observability and evaluation from day one
A Look Inside the Architecture
At a glance:
- Command Agent orchestrates tasks and coordinates workflows
- Domain Agents like Production Analyst, Forecaster, and Finance Analyst specialize in E&P workflows
- Chain of Thoughts enables structured reasoning, traceability, and better evaluation
- MCP Servers access PostgreSQL, Wolfepak, SharePoint, SCADA, and public APIs
- Azure Container Apps host each agent in scalable, isolated environments
- Azure AI Search, Redis, and Bing support retrieval, memory, and research
- Everything runs behind Azure Front Door, with authentication via Entra External ID and throttling through API Management
Secure by Design
PIXIE is built for use in regulated and high-stakes environments like oil & gas:
- User-level identity enforcement via Entra ID and JWT tokens
- Row-Level Security (RLS) at the database level for data isolation
- Secrets managed in Azure Key Vault
- Private endpoints for sensitive systems
- Full observability and auditing through Azure Monitor, Logs, and Guardrails
- LLM Evals + Logfire to monitor hallucinations and performance We built these into PIXIE because in energy, trust and traceability aren’t optional—they’re operational requirements.
What You Can Do With PIXIE
PIXIE agents are already powering workflows like:
- “Compare forecast vs actual oil production in Reagan and Loving Counties.”
- “Explain the drop in net revenue in Q2 by lease and working interest.”
- “What’s the impact on cash flow if Nymex oil drops to $60 next month?”
- “Find and summarize all SharePoint contracts mentioning ‘Thompson Unit’.”
PIXIE doesn’t just generate responses—it retrieves, reasons, and acts—across data, models, and documents.
Observability + Guardrails Built In
One of the most overlooked challenges in AI: debugging agent behavior. PIXIE includes:
- Logfire: Full trace logs of agent decisions, tool use, and outcomes
- LLM Evals: Test cases that validate agent chains and accuracy
- Azure Monitor: Track usage, performance, errors, and cost
- Guardrails: Enforce response structure, retry on error, and monitor quality
With PIXIE, AI decisions are visible, tunable, and auditable by your team.
Where We’re Headed
PIXIE currently supports a growing suite of agents. What’s next:
- Autonomous agents that schedule tasks, retry failures, and notify stakeholders
- Custom agents you can define with your own tools and data models
- Agent studio to plug in industry-specific copilots (e.g., LOE Tracker, Production Forecaster, AFE Reviewer)
We’re building toward a future where you can say: _“Generate next month’s cash flow using base case oil prices and send it to accounting.”
PIXIE makes that possible—securely, explainably, and at scale.
Final Words
PIXIE represents a new way of bringing AI into the oil & gas sector—not just as a chatbot, but as a transparent, orchestrated, agent-based platform that works alongside your team.
This isn’t a prototype. It’s deployed in production, improving every week, and tailored for the realities of energy.
If you're ready to put AI agents to work across your production, finance, or reporting workflows—reach out.
Let’s build AI that works like your best engineer, analyst, and teammate—24/7.