Soket AI, a Bengaluru-based AI research lab, has introduced LOOP, an open-source agentic harness that helps developers build and run AI agents while keeping control over their models, data, systems, and agent processes. The developer preview launched on October 8, 2026, for Linux, macOS, and Windows. Soket AI has been selected under the IndiaAI Mission to build foundation models for India and the Global South.
What LOOP Offers
LOOP is an infrastructure layer built in Rust for running AI agents on complex, long-running tasks. It focuses on low latency, efficient resource utilization, and parallel tool execution, and it supports skills and templates through slash commands. The platform also works with the Model Context Protocol, allowing developers to serve their own MCP tools while connecting models through a base URL.
Built for Long-Running and Multi-Agent Workloads
Soket AI designed LOOP for AI agents that must complete tasks over extended periods rather than ending with a single short interaction. Developers can branch, pause, and resume agent sessions while preserving session history during context compaction. In addition, multiple agents can collaborate on a task with shared memory, and optional isolation can be provided through rootless containers or microVMs.
Resource Efficiency Compared with Claude Code
In a local comparison conducted by Soket AI, LOOP and Anthropic's Claude Code were tested with the same Qwen model through an OpenRouter base URL and key. Both systems ran with eight workers, and the test measured only the harness process while excluding remote model inference. According to the company, LOOP consumed 7.7 times less RAM and 17 times less local CPU than Claude Code in the tested scenario.
Open-Source and Model-Agnostic Approach
LOOP is designed to give developers freedom over the AI models and systems they use. Organizations can link their own models, add MCP tools, and configure agent workflows according to their requirements, an approach suited to cybersecurity, banking, finance, and defense. Soket AI describes LOOP as the harness for systems you can control, enabling teams to avoid dependence on any single model provider.
Future Roadmap and Leadership
The current release is a developer preview, and several features are still in progress. Soket AI plans to add sandboxing, personas, multi-agent systems, layered memory, remote agents, trusted remote sandboxes, group chat for agents, tracking data, and a desktop application. Founder and CEO Abhishek Upperwal stated that a strong model cannot reach peak performance by itself and that the longer an agent runs, the more the harness matters.
Availability
LOOP is available as a developer preview on GitHub for Linux, macOS, and Windows. Soket AI is positioning the tool as an open-source alternative in the same broad category as agentic coding and execution tools such as OpenAI's Codex and Anthropic's Claude Code. Users can download the preview and test it across supported platforms.
With LOOP, Soket AI is addressing a growing need for AI infrastructure that can support long-running tasks while giving organizations control over their systems and data. The open-source release allows developers to inspect, modify, and integrate the harness into their own workflows. As the project evolves, its focus on resource efficiency, model flexibility, and secure isolation could make it a notable option for teams handling complex agent workloads.