Bengaluru-based AI research lab Soket AI has unveiled LOOP, an open-source harness designed to enable AI agents to work continuously on complex tasks for weeks rather than minutes. The developer preview is aimed at teams and institutions seeking greater control over the AI systems they operate, with applications spanning areas such as cybersecurity, banking, finance and defence.

Soket AI said LOOP is designed for environments where control and trust are critical. The company positions the product in the emerging category of AI agent harnesses, alongside systems such as OpenAI’s Codex and Anthropic’s Claude Code, while differentiating LOOP through its open-source architecture and model flexibility.

Built for Long-Running AI Workloads

As AI systems increasingly move beyond conversational interactions towards agents capable of taking actions, the infrastructure surrounding those agents becomes more important for longer-running tasks.

LOOP is designed to provide that infrastructure, allowing developers to use AI models from different providers by adding a model’s base URL. The harness then handles the rest of the integration.

The preview is built in Rust, with Soket AI highlighting its focus on low latency, resource efficiency and parallel tool execution.

Session Management and Multi-Agent Workflows

LOOP includes flexible session management that allows developers to branch, pause and resume sessions, with a workflow designed to resemble Git.

The platform also preserves session history during compaction, helping maintain context as tasks continue.

A single task can be distributed across multiple agents that share memory, while developers can optionally introduce isolation using technologies ranging from rootless containers to microVMs.

The preview also supports MCP (Model Context Protocol), along with the ability to turn skills and templates into slash commands and serve custom MCP tools.

Resource Efficiency

Soket AI has also published a local resource comparison between LOOP and Claude Code.

In the company’s comparison, both systems used the same Qwen model through an OpenRouter base URL and key, with eight workers. The measurement covered the harness processes only and excluded remote inference.

LOOP’s memory usage increased from 29 MiB to 223 MiB as workers were added, following a near-linear pattern. At eight workers, the company reported that LOOP used 7.7 times less RAM and 17 times less local CPU than Claude Code in the test setup.

Soket AI has clarified that these measurements relate only to local resource consumption and should not be interpreted as a quality, accuracy or task-success comparison.

Open Architecture for Developers

The company has developed LOOP’s core technology and infrastructure internally, with the harness designed to operate close to the system layer.

Developers can use their preferred AI model, connect MCP tools and configure agent workflows according to their requirements. The open-source architecture is intended to give teams greater control over how their AI systems are deployed and operated.

The company describes LOOP as “the harness for systems you can control.”

Next Layer Under Development

Soket AI said it is working on additional infrastructure for workloads where reliability and control are critical.

The company’s longer-term roadmap includes sandboxing, personas, multi-agent systems, layered memory, remote agents, trusted remote sandboxes and group chat for agents.

These capabilities are not part of the current LOOP preview. Telemetry and a desktop application are also still under development.

Available Across Major Operating Systems

LOOP is now available as a developer preview on GitHub for Linux, macOS and Windows.

The release gives developers an open-source environment for building and operating agentic AI systems while retaining control over the models and infrastructure they use.

Soket AI and IndiaAI Mission

Soket AI is a Bengaluru-based research lab developing AI models and systems that institutions can operate under their own control.

The company has been selected under the IndiaAI Mission to build frontier foundation models for India and the Global South.

At the opening ceremony of the India AI Impact Summit in New Delhi in February 2026, Soket AI founder and CEO Abhishek Upperwal was called on stage alongside global AI leaders including Sundar Pichai, Sam Altman and Dario Amodei.

With LOOP, Soket AI is expanding its focus from foundation-model development to the infrastructure required to operate increasingly capable AI agents.

Soket AI’s Agentic AI Push

Commenting on the launch, Abhishek Upperwal, Founder and CEO, Soket AI, said that strong models alone cannot reach their peak performance without the right harness, particularly as agents operate for longer periods.

With LOOP now available as an open-source developer preview, Soket AI is positioning the platform as an infrastructure layer for teams looking to build AI systems that can operate with greater flexibility, control and customisation.

By : Vanshika Tayal