Simulator Controller: An AutoHotkey Pit Crew That Talks to GPT Models

What Simulator Controller Actually Automates

The core of the project is described as a plugin capable automation framework that can drive external controller hardware such as button boxes, and through that hardware control simulator components like SimHub and SimFeedback. On top of that layer sit several voice chat capable assistants built on artificial intelligence technologies. The README names four: an AI Race Engineer that prepares pitstops, watches the weather forecast and estimates damage impact on lap times; an AI Race Strategist that adapts plans to race position, traffic and weather; an AI Race Spotter that warns about nearby cars; and an AI Driving Coach that answers handling questions and gives on-track instruction. The intended user is a sim racer who wants to keep both hands on the wheel and off the keyboard during a stint. It is not a telemetry overlay and not a game plugin in the usual sense. It is a Windows-side administration application that sits beside the simulator and talks to it through other tools.

The AutoHotkey Layer and Where the AI Lives

The primary language is AutoHotkey, and the README is explicit about why that matters: the language uses tricks to control Windows applications, tricks also used by malware, which is the stated reason antivirus products raise warnings. That single design decision explains most of the project's operational friction. The assistants are described as based on artificial intelligence technologies and the repository topics include gpt and llm, but the README does not spell out the request path from a spoken phrase to a model response. What can be confirmed from the material is that the assistants are voice chat capable, that they are separate roles with separate responsibilities, and that the framework around them is modular and extendable. Treat the assistant layer as a set of role-specific voice interfaces over shared race state, not as a single chatbot. The repository also lists stream-deck and streamdeck-plugin among its topics, so Stream Deck hardware is a first-class input path alongside button boxes.

Installing It Without Fighting Your Antivirus

There are two documented routes. The automated installer is a small executable downloaded from the project's S3 bucket, and the README notes you may have to deactivate antivirus or browser download protection to run it. It connects to the version repository and installs the latest version. The manual route is to download a release or beta archive and unzip it anywhere, then, beginning with Release 3.5.2, run the Simulator Tools application in the Binaries folder, which guides the rest of the installation. The README's own advice for persistent antivirus blocks is to define exception rules for two directories: the install directory and the one under the user's Documents folder. If exceptions are not possible, the README says plainly to look for another tool. The current stable build listed is 7.2.3.0-release, which migrated to AHK 2.1-Alpha.31 and added support for ACE setup files. A 7.2.4.0-beta development build is also published. The release notes and update notes are separate documents, and the README asks you to read the update notes for every version you skipped, not just the newest one.

The Assistant Roles Are Not Interchangeable

Reading the feature list as one product undersells how differently the four assistants behave. The Race Engineer is reactive and task-oriented: pitstop preparation, weather, damage estimates. The Race Strategist is stateful across a whole race and adjusts to position, traffic and weather changes, which means its usefulness depends on how much race state the framework can actually observe. The Spotter is latency-sensitive in a way the others are not; a warning about a nearby car that arrives late is worse than no warning, and the README gives no timing figures. The Driving Coach is the least time-critical and the most open-ended, answering questions about car handling and real-world racing. If you adopt the project for one role, evaluate that role alone. A setup that works well for pitstop preparation tells you nothing about whether the Spotter responds fast enough during a close fight.

Release Cadence and What Upgrades Cost You

The 7.2.x line moves quickly. Between 7.2.1.0 and 7.2.3.0 the changelogs cover translations, telemetry handling, plugin registration method, sound playback workarounds, tyre change logic, car metadata for AMS2, LMU and R3E, and database synchronization filters for unknown cars and tracks. Several entries are fixes for crashes or for features that simply did not work, such as repair requests for LMU and lap-based tyre change. That pattern means upgrades are not cosmetic. Skipping versions accumulates update notes you are expected to read before installing. The practical cost is not the download; it is re-verifying your controller bindings, voice options and simulator-specific data after each jump, because the changelog shows those surfaces changing. The repository's licence is reported as NOASSERTION, which means no standard licence identifier was detected. The README mentions donations and a Patreon tier that grants access to a public Team Server for multiplayer endurance races. What that tier does and does not grant is not stated in the material available here, so check the repository's licence file and Patreon terms yourself before relying on any of it commercially.

Where It Is the Wrong Tool

If your machine is managed by an IT policy that forbids antivirus exceptions, this project is effectively unusable, and the README says so rather than pretending otherwise. If you race on a console or on a platform where you cannot run arbitrary Windows applications alongside the simulator, the automation framework has nothing to attach to. If you want a lightweight overlay that shows tyre temperatures and fuel, the scope here is far larger than that and the installation burden is disproportionate. There is also a support-surface problem: the README points to a wiki, release notes, update notes, a video and forum threads, but no single document describes the AI request path or the failure behaviour when a model call fails mid-race. For an assistant that speaks to you while you drive, that is an unanswered question worth resolving before you depend on it in a league race.

Compared With Crew Chief and Similar Voice Tools

The obvious comparison class is voice spotter and race engineer tools such as Crew Chief, which generate spoken callouts from simulator telemetry using rule-based logic running locally. The difference in approach is the source of the words. A rule-based spotter decides what to say from fixed conditions and says it the same way every time, with predictable latency and no external dependency. Simulator Controller routes at least part of the interaction through AI assistants, which the README ties to GPT technology, so responses can be phrased variably and can answer open questions such as the Driving Coach's handling queries. The trade is dependency and unpredictability: an external model call introduces a network path and a response time that a local rule engine does not have. If you want deterministic, offline callouts, the rule-based approach is the better fit. If you want to ask questions mid-session and get an answer in natural language, that is the gap this project is aiming at.

Editorial conclusion

Adopt Simulator Controller if you already run SimHub or SimFeedback, want voice-controlled pit and strategy workflows, and are willing to configure an external AI backend and whitelist its folders in your antivirus. Do not adopt it if you want a zero-configuration overlay or cannot tolerate AutoHotkey-based tools on a locked-down machine. Before installing, verify which AI provider and API key the assistants require, confirm your target simulator appears in the supported car and track databases, and read the Update Notes for every release you skip, since the 7.2.x line ships frequent data and behaviour changes.

Official sources

1. Issues
2. README
3. Releases
4. SeriousOldMan/Simulator-Controller on GitHub