GAMESLASHNPCTalk releases AI-agent toolkit to make game NPCs answer spoken dialogue
By GameSlash
Open-source NPCTalk connects player speech to game NPCs through an LLM and Fish Audio, with a GTA V mod example and documented external service costs.
NPCTalk is an open-source collection of agent skills and voice-server tools for making NPCs respond to players' spoken words. The developer first built a GTA V story-mode example, while separating the core components so agents can adapt them to other games through mods or game APIs.
From the player's voice to an NPC response
The documented pipeline takes microphone input, transcribes it, sends NPC identity and scene context to an LLM, then uses Fish Audio Drama 3 or S2 Pro to produce voice alongside subtitles, mouth movement and character actions. The developer reports 1.2–2.2 seconds of end-of-speech-to-audio latency for push-to-talk; GameSlash has not independently reproduced this number.
The GTA V example must provide game-specific hooks for identifying the NPC, reporting the scene and exposing allowed actions. Other games require equivalent custom integration: installing NPCTalk alone does not make NPCs in every game speak.
Multiple supported coding agents, external running costs
The repository describes installation routes for Claude Code, Codex, Cursor, Gemini CLI, Copilot and OpenCode. Its npctalk CLI checks server health, prepares voice references and generates test audio. Live operation depends on external infrastructure, including Modal, fal and an LLM provider, which can incur charges. Cloning a voice also requires authorization for the source recording.
NPCTalk is a modding/developer toolkit, not an official GTA V or Rockstar feature. GameSlash reviewed the source documentation and diagrams, not the installed mod. Performance and compatibility will depend on each game's integration and hardware.
Sources: NPCTalk repository | MIT license