BeyondATC, the AI-powered ATC addon for Microsoft Flight Simulator, has reached significant milestones in its development this month. Last week, it introduced advanced AI traffic injection capabilities, and now the team has announced the integration of a custom-built Large Language Model (LLM) into its framework. This update aims to enhance user interaction with the system, addressing long-standing challenges and introducing new opportunities for simmers.
From its inception, BeyondATC aimed to provide a more immersive and realistic ATC experience, stepping beyond the limitations of Microsoft Flight Simulator’s default system. While artificial intelligence has always been at the core of its operations—from voice synthesis to logic processing—the absence of an LLM was a deliberate choice in earlier iterations. In their announcement this weekend, the developers cited two main hurdles: functionality and cost.
Initially, LLMs like OpenAI’s ChatGPT were unsuitable for the task. They lacked the structured context necessary for ATC operations, often “hallucinating” nonsensical instructions or responding inaccurately to user inputs. BeyondATC’s developers instead focused on building a structured “on rails” system to ensure all ATC interactions remained realistic and within the bounds of aviation protocols. This groundwork has now paved the way for the introduction of an LLM that works harmoniously within the program’s robust framework.
The high cost of running commercial LLMs was another obstacle. BeyondATC’s solution? A homegrown LLM specifically trained for its unique 3D engine and simulation environment. This approach not only enables new capabilities on the platform but also dramatically reduces operational costs.
As a result, the upcoming LLM integration will be included as a free update for all users, with no additional subscription or usage fees—a decision that aligns with the program’s goal of being an accessible tool for simmers.
What does the LLM bring to the table?
The integration of an LLM introduces dynamic capabilities that enhance the realism and flexibility of ATC communications in the sim, a concept that we’ve seen working very well already in BeyondATC’s closest competitor, SayIntentions.AI. There are several key advantages made possible by this:
- Natural Language Interaction:
- Simmers can communicate with ATC in a conversational manner, asking detailed or open-ended questions about their flight, the airport, or surrounding traffic.
- Examples showcased in the video accompanying the announcement included queries about active runways, distances to nearby traffic, and even whimsical questions—all handled seamlessly by the LLM.
- Situational Awareness:
- The LLM draws from BeyondATC’s internal 3D engine to understand the user’s flight environment. This aims to ensure that responses are accurate and contextually appropriate.
- Local Processing:
- Unlike many LLM-based tools, BeyondATC’s model runs locally on the user’s machine. This design choice reduces dependency on external servers, mitigates costs, and ensures tighter integration with the ATC framework.
- Improved Accessibility:
- For users who prefer not to use voice commands, a new on-screen keyboard allows typed interactions with ATC.
Building toward VFR capabilities
This update also lays the foundation for BeyondATC’s future development. The integration of an LLM is a critical step toward supporting VFR, which typically entails more open-ended and less structured communications. With the groundwork now in place, the developers can shift focus to this ambitious goal, bringing even greater realism and flexibility to the platform.
Early access and roadmap
The new BeyondATC with LLM-powered functionality will be released in early 2025. As with previous updates, it will first be available to supporters through an experimental branch before rolling out to all users as a free update. This measured approach will allow the team to fine-tune the system based on community feedback, hopefully ensuring a polished final release.
