Via
Via is a middleware layer that sits between enterprise CLIs/IDEs and LLMs to cut AI coding costs by up to 67.4% while maintaining or improving accuracy. It installs in one terminal command, runs entirely within the enterprise's VPC, and integrates with tools like Claude Code, Codex, and Cursor. The platform is powered by the Umami agent paradigm and Codegraph compression algorithm.
- Developer Tools
- Artificial Intelligence
- AI Agents
- Software Only
Funding
Funding not disclosed
Founders
Founder details are not available yet.
Product
Problem
Enterprise AI systems struggle to achieve both high accuracy and cost efficiency in agentic coding tasks, where token spend is exponentially growing and existing solutions either compromise coding accuracy or disrupt the developer experience. By 2028, Gartner projects AI coding costs will exceed the average developer's salary, making this a critical operational challenge.
Solution
Via provides a middleware layer that sits between a CLI or IDE and an LLM, routing requests through an enterprise's VPC to reduce token consumption without sacrificing output quality. The platform integrates with existing developer tools like Claude Code, Codex, and Cursor, installing in roughly ten seconds with one terminal command. Via employs the Umami agent paradigm—an in-house model-agnostic agent framework built on graph and information theory research—to optimize how coding tasks are processed and compressed. The company has demonstrated a 67.4% cost reduction on DeepSWE 1.1 at equal accuracy to Claude Code, and a 33.5% cost reduction with higher accuracy on Terminal Bench 4. No data ever leaves the enterprise environment, addressing security and compliance requirements.
Target Audience
Enterprise organizations with seven-figure AI coding budgets, including large technology companies and regulated industries like healthcare that require data to remain within their infrastructure.
Features
- Middleware architecture that sits between CLI/IDE and LLM, compatible with Claude Code, Codex, Cursor, and other development tools
- One-command terminal installation with full IDE and CLI integration in under ten seconds
- Umami agent paradigm built on graph and information theory research for LLM-optimized compression (Codegraph algorithm)
- Role-based budget controls, spend observability, and forecasting dashboards for finance teams
- VPC-deployed infrastructure ensuring no data leaves the enterprise environment, SOC 2 Type II certified
- Achieved #1 ranking on SWE-Bench Verified for 5 months of 2025 as the most accurate coding harness (single model, no test-time compute)
- Works with both public and internal gateway configurations