Veea Open-Sources Lobster Trap and Partners with NativelyAI to Advance Secure Agent Deployment

Free, open-source software inspects every conversation between AI agents and the models they rely on. It is integrated within NativelyAI’s 250,000+ developer ecosystem and available as part of TerraFabric, Veea’s control plane for governed autonomous systems at the edge

Barcelona, Veea announced the open-source release of Lobster Trap, a lightweight security tool that monitors and enforces rules on interactions between AI agents and the language models that power them.

To accelerate enterprise adoption and embed conversation-layer security directly into development workflows, Veea is partnering with NativelyAI’s builder community platform, lablab.ai, which has more than 250,000 AI developers building and deploying AI applications. Through this collaboration, Lobster Trap will be packaged within Native.Builder, NativelyAI’s AI software production platform, enabling development teams to deploy AI agents with policy enforcement built in.

The partnership places Lobster Trap inside an established AI builder ecosystem and accelerates the development of enterprise policy packs, reference integrations, and secure deployment templates for production environments.

Also Read: AiThority Interview With Arun Subramaniyan, Founder & CEO, Articul8 AI

The Problem: AI Agents Are Getting Access. The Guardrails Haven’t Kept Up.

AI agents are increasingly given the ability to read files, write code, send messages, and take actions inside real business systems. That capability introduces significant risk to enterprises. A manipulated prompt or an unexpected model response can expose passwords, leak sensitive data, or trigger unintended actions.

Most organizations today have no visibility into what their AI agents are asking a model to do, or what the model is responding with in return. Existing web and API security tools were not designed to inspect this conversational layer.

“The industry has spent the last two years racing to give AI agents more power,” said Allen Salmasi, Founder and CEO of Veea. “What has been missing is a practical way to observe and enforce policy at the point where the AI agents interact with AI models. Lobster Trap addresses that gap.”

How It Works

Lobster Trap runs inline between AI agents and the language models they communicate with. Every prompt sent by the agent and every response returned by the model is evaluated against defined security policies before the agent is allowed to proceed.

If a violation is detected, Lobster Trap can block the interaction, flag it for review, or log it for analysis.

The scanning occurs under a millisecond and introduces no meaningful delay. The tool works with AI backends that use the standard OpenAI-compatible interface, allowing most deployments to adopt it without modifying application code.

Security policies are defined in a configuration file. Out of the box, Lobster Trap detects prompt injection attempts, credential exposure, personal information leakage, suspicious file access, and data exfiltration patterns.

NativelyAI Partnership: Enterprise Adoption and Secure Agent Templates

Through its collaboration with NativelyAI and its community product LabLab.ai, Veea is supporting enterprise policy packs, reference integrations, and secure agent blueprints that make conversation-layer controls repeatable across teams and industries.

By packaging Lobster Trap inside Native.Builder, development teams can launch agent-based applications with policy enforcement enabled by default. This reduces the need to retrofit controls later and simplifies secure deployment in enterprise environments.

Why TerraFabric

TerraFabric is Veea’s control plane for governed autonomous systems at the edge. It manages distributed infrastructure as coordinated systems rather than isolated devices, applying policy enforcement, orchestration, and lifecycle control across fleets.

Lobster Trap extends that governance model to the conversation layer. Where TerraFabric governs what workloads run and where they run, Lobster Trap governs what those workloads are allowed to ask and what they are allowed to receive from language models.

In a TerraFabric deployment, Lobster Trap runs on local hardware. Prompt data, model responses, and audit logs can remain on-premises, supporting security and data sovereignty requirements.

“Autonomy without control introduces risk,” Salmasi said. “TerraFabric governs infrastructure. Lobster Trap governs AI conversation. Together, they give operators policy enforcement at every layer.”

Open source by design

Veea is releasing Lobster Trap under the MIT license, enabling developers to use, modify, and extend it freely. The project is written in Go and compiles to a single file with no external dependencies, allowing it to run on Linux, macOS, or Windows.

Also Read: Cheap and Fast: The Strategy of LLM Cascading (Frugal GPT)

[To share your insights with us, please write to psen@itechseries.com]