AI Dev 26 X SF | Aman Singla & Aseem Chandra: MarcoPolo, A Workspace for AI to Work with Your Data

Why It Matters

Marco Polo turns fragmented AI‑to‑data integrations into secure, reusable workspaces, unlocking enterprise analytics for non‑engineers while preserving data governance.

Key Takeaways

• Marco Polo creates secure AI workspace linking LLMs to enterprise data.
• Unified CLI abstracts 50+ data sources, reducing LLM context load.
• Persistent workspaces store schema, queries, and audit trails for reuse.
• Scoped credentials enforce security, preventing direct LLM data access.
• Contextual learning enables agents to join data across systems efficiently.

Summary

The video introduces Marco Polo, a middleware platform that provides a dedicated workspace where agentic AI models—such as Claude, ChatGPT, or custom agents—can safely access and manipulate enterprise data across dozens of systems. By running in a secure Kubernetes container, Marco Polo bridges LLMs with raw storage, databases, CRMs, ERPs, and support tools while enforcing architectural boundaries and credential scoping.

A core insight is the unified command‑line interface that abstracts over 50+ data sources, offering verbs like list, query, and upload. This reduces the prompt length and token consumption needed for the LLM to understand each source’s API. The platform also pre‑loads schema and connection metadata, solving the “cold‑start” problem and allowing the AI to generate accurate SQL or JQL queries from day one. Persistent workspaces retain query history, schema, and audit logs, enabling the model to reuse prior work and progressively improve its context.

A concrete demo shows an ops user asking for churned customers and their support tickets. Marco Polo supplies the correct Salesforce object, joins it with Jira tickets via a temporary DuckDB instance, and even builds a rerunnable revenue‑by‑region dashboard by stitching together Salesforce and census data. The LLM never sees raw credentials; instead, it invokes privileged CLI calls that enforce scoped access.

The implications are significant: non‑technical staff can leverage powerful LLM agents without writing code, enterprises gain a searchable, auditable knowledge base, and security teams retain control over data exposure. By turning ad‑hoc data pulls into repeatable, token‑efficient workflows, Marco Polo aims to democratize AI‑driven analytics at scale.

Original Description

AI assistants like Claude, ChatGPT, and Cursor are transforming how we work, but working with real data remains a challenge.

Secure Access: How does AI safely reach your databases, APIs, S3, CRMs, and other systems?

Capability: How can it query, transform, and analyze — not just read?

Context: Which data matters and how should it be used?

Aman Singla & Aseem Chandra showed at AI Dev 26 x San Francisco how MarcoPolo solves this with secure, persistent workspaces where AI can work with your actual data across 50+ systems — without exposing credentials to the model.