Article · 9 November 2026

Voice and LLM strategy games in 2026: who does what

Written by the dev team’s AI helper

We searched for every strategy game that you command by voice or that uses a large language model (LLM), and read their store pages. The short version: there are two families of games, and they almost never overlap.

Family one: voice games with fixed phrases

These games listen for commands they know in advance. Recognition is fast and reliable, but you speak the game's language, not yours. None of them uses an LLM to understand you.

• Tom Clancy's EndWar (2008): a fixed grammar, unit, action, target. The voice system was praised; the campaign was not.
• Radio General (2020): Second World War, orders over the radio, English only and sensitive to accent. You never see your troops, only their reports, which can be confused. Veteran companies carry over between missions.
• Radio Commander: Vietnam '64: a scripted campaign driven by radio reports, with voice recognition on Windows.
• There Came an Echo (2015): squad tactics designed around voice, but fully playable without it.
• Speak of The Crown (release date to be announced): a fantasy RTS played by voice, English only.
• Castle Siege Commander (early access planned): medieval sieges and online PvP, with voice commands whose phrases you define yourself, in 13 languages.

Family two: LLM games, mostly turn-based

These games let you write, and sometimes say, anything; a language model interprets it. Almost all of them are turn-based or grand strategy, because a model needs time to think.

• Command Doctrine (free, 2026): turn-based battles and campaigns generated from a prompt; orders by voice or text, interpreted by Gemini online.
• Command by Voice: World Conquest (early access, 2026): grand strategy run by voice, Chinese and English, using DeepSeek in the cloud.
• ARBO: Arena Tactics: a turn-based sci-fi tactics game where an AI carries out free-text orders and may disagree with you; cards shape its personality.
• DIRECTIVE (open source): Barbarossa 1941, you write directives to AI generals with temperaments who may soften or ignore them. It runs on a local LLM through LM Studio, Ollama or llama.cpp, but one turn takes minutes.
• Arma Reforger mods such as LLM Commander and Stavka: text orders to AI squads through a language model.

The gap between them

Put the two families side by side and one combination is missing: a real-time battle, commanded by free speech, understood by an LLM that can run on your own computer. Voice games are real-time but cannot understand free speech. LLM games understand free speech but stop the clock to think.

That gap is what Adjutant is built for:

• Real time. The battle does not pause. You speak, the adjutant answers within seconds, and the simulation moves the soldiers.
• Free speech. No phrase list. "Keep the spears on the hill and send the horse round the wood" is a valid order.
• A local LLM. Out of the box a local open model (Qwen) runs with llama.cpp on an 8 GB graphics card, offline. Any OpenAI-compatible server works too, including Ollama and LM Studio; or use Claude or ChatGPT.
• Speech plus the map. Draw arrows and zones, give them names, and use them in your orders. No other game here combines the two.
• A character, not a parser. The adjutant reports back out loud, and tells you when an order makes no sense.
• PvP through two AIs. Each player commands through their own adjutant.

How to read this list

If you want a polished, proven voice game today, Radio General is the safest pick. If you want to experiment with LLM command, DIRECTIVE and Command Doctrine are the most interesting. If you want to speak freely in a real-time battle and have an AI officer carry out your plan, that is what we are making.