Modus lands $10M to equip AI agents with precise, up-to-date business context

Deploying AI agents across enterprise systems raises a pressing challenge: how do teams ensure these agents truly grasp how the business actually works? Many organizations today resort to manually crafting and maintaining Markdown documentation, a practice that quickly becomes stale as operations evolve.

Modus, which exits stealth this week with a $10 million seed round, is tackling this problem with what the industry calls a context warehouse—a specialized layer that operates alongside conventional data infrastructure. The platform continuously monitors how business operations function across internal systems, then delivers only the relevant information to an AI agent at the precise moment it's needed.

How Modus mines and composes context#

The startup ingests data from tools like GitHub, dbt, Jira, Snowflake, and Postgres using its Context Miner, which runs continuously to build a living map of business operations. Co-founder and CTO Tomer Mesika explains that the system determines autonomously what to extract, how frequently to check, and when to investigate deeper.

The mined information then flows through the Context Composer, a real-time assembly system that generates what Modus calls "dynamically generated skills"—concise, purpose-built briefs tailored to each specific task. These briefs reach the agent ready-made, eliminating wasteful pre-work.

Co-founder and CEO Daniel Shimoni frames the opportunity by drawing a parallel to data warehousing. "There's a logic behind data warehouses," he notes, but organizations lack equivalent infrastructure for managing organizational knowledge and context. The bigger challenge isn't building context once—it's keeping it current as the business shifts. Modus addresses this by continuously learning from operations and refreshing only the relevant context whenever change occurs.

Targeting the cost crisis#

Enterprise AI budgets face mounting pressure, with token consumption emerging as a central cost driver. Modus pitches a solution: smaller, cheaper models currently waste effort sifting through documentation to determine relevance before expensive frontier models even begin their work.

Rather than perform this triage on-demand, Modus uses smaller language models, vector search, and graph databases operating continuously in the background to prepare finished briefs in advance. By the time a costly frontier model engages, it receives exactly what it needs—no more context-hunting overhead, no token waste on menial tasks.