SandboxAQ Launches Switch to Coordinate Multi-Framework AI Agents in Enterprise Chat

SandboxAQ has introduced Switch, a source-available coordination platform designed to integrate heterogeneous AI agents directly into standard enterprise communication channels, including Slack, Microsoft Teams, and Discord. Part of SandboxAQ's broader Flint AI ecosystem—which includes the Flint CLI for local agent evaluations—Switch provides shared virtual "rooms" where humans and automated agents share conversation histories, context, and resources. The tool supports agents built across differing frameworks such as Claude Code, Google Agent Development Kit (ADK), LangChain, and OpenAI via Model Context Protocol (MCP) integrations, APIs, and modular adapters under an Apache 2.0 license with Commons Clause restrictions. For platform engineers, DevOps leads, and SREs, this move targets the growing friction of managing isolated agent silos. As engineering organizations deploy multiple task-specific agents—for automated pull request reviews, infrastructure drift detection, security triaging, and documentation generation—human engineers frequently become human middleware, copying prompts, logs, and outputs between separate agent dashboards. By standardizing agent execution within existing collaboration fabrics, teams preserve shared state across complex tasks without maintaining ad-hoc integration code for every new model or tool harness. This release reflects a pivotal architectural shift in enterprise AI agents. The industry is transitioning from single, standalone conversational assistants to multi-agent, cross-functional ecosystems that must interact with both human teams and third-party systems in real time. As standard protocols like MCP gain traction, the key differentiator for agentic productivity is no longer solely the underlying model’s raw reasoning capability, but how effectively context and state persist across multi-step human-in-the-loop workflows. Embedding agents where operators already coordinate reduces handoff latency and simplifies auditability. In practice, engineering teams evaluating Switch should assess how chat-based agent execution intersects with their identity and permission boundaries. Running multi-agent workflows across production chat channels requires careful scoping of agent credentials, strict least-privilege tool access, and robust rate-limiting to prevent cascading automated loops. Additionally, while the core software is freely accessible for internal enterprise deployments and development, organizations building commercial solutions must note the Commons Clause commercial resale restrictions. Teams should start by piloting agent rooms on constrained internal workflows, such as post-incident timeline reconstruction or release staging, before expanding agent privileges into automated production changes.