Vision · Pre-release

Why we're building mantle

Mantle is the context layer for AI agents — connecting data, resolving entities, and serving quality-scored context to any agent through a single API.

AI agents are flying blind.

Today's AI agents are remarkably capable — they can reason, plan, and execute complex workflows. But they share a critical weakness: they lack context. Enterprise data is scattered across dozens of tools, siloed in formats that agents can't access, and impossible to query in real time.

The result? Agents that hallucinate, make decisions on incomplete information, and require constant human intervention. The bottleneck isn't intelligence — it's context.

mantle is the missing layer.

We're building mantle as a semantic context layer that sits between your data and your AI agents. It is being designed to connect to your data sources, build a real-time knowledge graph with entity resolution, and serve quality-scored, token-optimized context via the Model Context Protocol (MCP).

One API call. Every connected source. Quality-scored context.

What we're building

Zero-Copy Connectors

We're building connectors that read your data in place — CRM, data warehouse, documents, chat, email — without ETL pipelines, sync jobs, or data duplication. Your data stays where it lives.

Semantic Knowledge Graph

Automatic entity resolution and relationship mapping across connected sources. mantle is being built to understand that "John Smith" in Salesforce and "jsmith@acme.com" in Slack are the same person.

Quality-Scored Context

Every piece of context is intended to be scored for relevance, freshness, and reliability — so agents receive signal, not noise.

MCP-Native Delivery

Built on the Model Context Protocol from day one, with the goal of token-optimized payloads delivered to any MCP-compatible agent framework.

Security by Design

We're designing mantle around a zero-copy approach so sensitive data does not need to leave its source system. See the security page for our roadmap.

Built for developers.

Five lines of code to add deep context to any AI agent. We're building SDKs for TypeScript, Python, and Go.

// One API call — any language
const ctx = await mantle.semantic_search(
"Q3 revenue by region"
);
→ resolved entities · ranked sources · quality-scored context
Illustrative — actual numbers vary by data and connectors.

What's next.

This is just the beginning. We believe the context layer will become as fundamental to AI infrastructure as the database is to applications.

Mantle is live today. Sign in to create a project and connect your first source — most teams are querying their own data within minutes.

Ready to give your AI agents real context?

Get started in five minutes.

The mantle team