Moderne launches Trigrep to speed enterprise AI agents

Moderne has added a code search feature called Trigrep to its Agent Tools platform, aiming to speed up coding agents working across large codebases and multiple repositories.

The launch comes as more organisations move coding agents from experimentation into production workflows. As a result, execution efficiency is drawing more attention: broad scans across entire code estates can add latency and increase compute usage at scale.

Trigrep sits alongside Moderne's existing tools for agent-driven software engineering. It is designed to reduce how much code an agent must inspect before deeper analysis or automated changes begin.

How Trigrep works

Trigrep uses trigram indexing compatible with Zoekt, an open source code search engine. It builds its index from OpenRewrite's Lossless Semantic Trees, which represent code with compiler-level accuracy.

Rather than indexing raw source files, Trigrep indexes the source and structural information captured in this semantic representation. Searches then run against precomputed trigram indexes.

This design is intended to keep discovery fast across large repository estates because search performance depends on pattern matches rather than total repository size. Moderne says it enables sub-second discovery when agents look for relevant files, functions, or patterns before running further workflows.

In practice, Trigrep is meant to reduce the number of files or repositories that need semantic analysis before automation proceeds. Moderne says this lowers compute overhead and speeds up downstream analysis and transformation.

Execution layer

Moderne positions Trigrep as execution infrastructure for coding agents rather than a developer browsing tool, calling it an "execution primitive" inside its control plane.

It supports literal searches, boolean queries, regular expressions, and structural pattern matching. It also persists search results for later workflows, allowing an agent to reuse a previously identified scope instead of repeating broad discovery.

Trigrep is described as agent-agnostic and designed to fit into AI-driven workflows in enterprise environments. Within the Moderne platform, it becomes part of an orchestration layer that operates across organisation-scoped sets of repositories, with governance applied to execution.

Interest in these layers has grown as companies try to manage operational risks from automated code changes, including unintended modifications and inconsistent rollouts. Compute cost has also become a constraint for teams running agent-based workflows across large software estates, especially when repeated scanning makes workloads expensive and slow.

Moderne argues that the shift to production use changes the bottlenecks for coding agents. Model quality affects what an agent can propose, while execution performance determines whether those proposals can run across a large portfolio without long queues, heavy infrastructure use, or failures from timeouts and resource limits.

Jonathan Schneider, CEO and co-founder of Moderne, tied Trigrep to this broader shift in priorities for agent-driven engineering.

"The infrastructure that optimizes how agents execute is becoming as important as the models themselves," said Jonathan Schneider, CEO and co-founder of Moderne. "Trigrep delivers sub-second discovery that narrows the execution surface before automation begins. At enterprise scale, speed and precision are what keep agent-driven engineering efficient."

Availability

Trigrep is available as part of the Moderne command line interface, which is used to run workflows against codebases managed in the platform.

Moderne is based in Miami and counts Acrew Capital, Intel Capital, True Ventures, Mango Capital, Allstate Strategic Ventures, Morgan Stanley Ventures, Amex Ventures, and TIAA Ventures among its investors.

Moderne expects execution infrastructure to become a more prominent layer as organisations operationalise coding agents across large software portfolios.