Complete guide

Invite a character along and they walk at your shoulder through Night City. Ask them something out loud, into your own microphone, and they answer in their own voice: lips moving with the words, eyes on you, hands talking when the story runs long. Sit down on a couch and they sit beside you. Stay a while in your apartment and they find a seat or a spot to stand instead of trailing you around the flat. Get in the car and they ride along, or take the wheel if you ask. Later, text them from V's phone, or ring them on a holocall and watch them pick up.

AI Companions (LLM) gives the game's named characters, from Panam and Johnny to Takemura, Kerry, River and Alt Cunningham, personalities researched from the game's own lore, memories that last, and voices that sound like the characters you know. And the street answers too: the ganger on the corner, the NCPD officer, the ripperdoc and the bartender each talk back as their kind of person.

Free, three ways#

• Fully local and free. Download a Local AI pack, extract it into the game folder, and your characters think and speak entirely on your own PC (no server or other app to set up), with the Pocket TTS voices: no account, no key, no internet, no limit (only error reports, if you leave them on, go out when something fails). The smaller Local AI model keeps smaller graphics cards in the game.
• Google Gemini, the free cloud option. Fast replies, about a second to the first words, and light on your PC. Make a free key at Google AI Studio (a Google account is enough, no card), choose Google Gemini (API key) in AI provider, press Add and you are done (details). Unlimited on your own key: your first thousand requests each day go to Google’s Flash Lite models, then Gemini Live takes over. It is not offered in every country, and Google may use free-plan chats to improve its products.
• OpenRouter, one key. Make a key at openrouter.ai/keys (a free account is enough, no card), choose OpenRouter (your key) in AI provider, press Add and you are done (details). The mod picks a free model that can see pictures. A free key gives about 50 requests a day (roughly 20 conversation turns). Buying 10 dollars of credit once raises it to 1,000 free requests a day, and free models never spend that credit. For another model, use a Custom endpoint.

Mistral. Mistral used to give everyone a free API key with about 5,000 turns a day, but new accounts don't seem to get one any more. Still worth a try for that allowance and its Voxtral voice cloning: it is in the provider list again as Mistral (your key), with its own Mistral API key row on AI provider (details).

The voices are free too. The default, Pocket TTS, gives every character a voice that sounds like them in the game, running right on your processor, so a fresh install sounds like Night City from the first line. With a Mistral key (added on the Voxtral key row or the Mistral API key row), Voxtral does the same in the cloud.

---

Install#

Close the game. Extract each download into your Cyberpunk 2077 folder (the one with bin, archive and r6) and let the folders merge. If Windows asks about files with the same names, choose Replace: it is always safe. Install the mods under Requirements first.

Every download is a file on the mod's Nexus Files tab: the main files (the full install, Pocket Voice, Speech to Text (English), Local AI, Mature and OmniVoice) and the optional files (the two Ultra packs, the two other Speech to Text packs and the eleven Pocket Voice language files). Nothing is on Hugging Face except the OmniVoice server for another PC and the voice clips the mod downloads by itself. Each pack is one zip file, so there are no parts to join, and each works on its own: install none, some or all. Take the full install at every release and extract it over your old one to update (there is no separate update download), and the Pocket Voice and Speech to Text (English) files once (they have their own versions and are not downloaded again on a mod update). Extract the full install before the OmniVoice pack. If Windows' built-in extractor complains, use 7-Zip. If a pack ends up one or two folders too deep (for example inside a folder named after the zip), the mod moves its files into place the next time the game starts.

The first start after installing takes a little longer while the service prepares itself.

Update. By hand, close the game and extract the new full install over your install. Your conversations, memories, settings and optional packs (the Pocket voice file too) stay. One zip, no steps to apply in order. With Vortex or Mod Organizer 2, install the newer full install as your one mod instead (below). Choose Replace if Windows asks, which is always safe. Never delete the mod's folders to update: the optional packs live inside them. Files the new version no longer needs are removed by the mod itself at the next start, so you never delete anything. The small Runtime update zips of earlier versions (the updates folder, once on the Hugging Face page) are not offered any more. If an older version kept the service's data inside the mod's folder, the first start moves it to the place below.

Vortex. Add the full install (AI-Companions-LLM-<v>.zip) as one mod: drop it onto the Mods page, then click Deploy (after the first start Vortex may report "external changes" in this mod, where the mod copied its code: click "Save all changes", never Revert). It holds the Runtime too, so it is the only mod you need. To update, add the newer full install the same way and choose "Replace the existing mod" (about 0.55 GB, one click). Coming from 0.3.3 or older with a Scripts mod and a Runtime mod? Remove both, then install the full install. For the Local AI and Mature packs Vortex shows "You Have Reached The Fallback Installer": choose "Yes, Install To Staging Anyway". Two packs at once give "unresolved file conflicts": pick either one to load after the other. The OmniVoice pack must be extracted by hand (it holds DLLs). The warning sign in the Version column and "external changes" at deploy are harmless.

Mod Organizer 2. Install the full install as one mod. To update, install the newer full install the same way and choose Merge if asked, so the packs in its folder stay. Coming from 0.3.3 or older with a Scripts mod and a Runtime mod? Remove both, then install the full install. Every pack can be a mod of its own (from 0.5.3): Pocket Voice, each Pocket Voice language file, Speech to Text, Local AI and Mature (each with its vision file inside) and OmniVoice. If Mod Organizer 2 says a mod with that name already exists, choose Rename or Merge, never Replace, which removes the main file. Tick every mod in the left pane, the order does not matter, and start the game with Mod Organizer 2's Run button.

Coming from the separate Scripts and Runtime files (0.3.3 or older)? Since 0.3.4 the full install holds both, and Vortex or Mod Organizer 2 cannot know that it replaces them, so it adds a third mod beside the old two and nothing is replaced. Remove the old Scripts mod and the old Runtime mod first (in Vortex, Remove on the Mods page, in Mod Organizer 2, right-click and Remove Mod), then install the full install as one mod and deploy. Your memories, saved keys, custom characters and settings are kept: the memories, keys and characters live in %LOCALAPPDATA%\AiNpcCyberpunkMod, outside any mod, and the service keeps a copy of your settings there and puts it back at the first start. Anything you put into the old Runtime mod's own folder, such as the multilingual packs, goes with that mod, so put it into the new mod again. Packs that are mods of their own stay.

Where your data is. The background service keeps the memories stored on your computer, your saved keys, its logs and your reports in %LOCALAPPDATA%\AiNpcCyberpunkMod, outside the game folder. Your settings and Phone Chat history stay in the mod's own folder.

Uninstall. Delete these from your Cyberpunk 2077 folder, and keep the frameworks if other mods use them:

• bin\x64\plugins\cyber_engine_tweaks\mods\AiNpcCyberpunkMod (this also deletes your settings and Phone Chat history)
• red4ext\plugins\AiNpcCyberpunkMod (this also removes the OmniVoice, Local AI and Mature packs)
• red4ext\plugins\AiNpcCyberpunkModNative
• r6\scripts\CyberpunkConvai (the scripts; internal name), and r6\audioware\CyberpunkConvai if an older version left it
• archive\pc\mod\convai_facial.archive, convai_gestures.archive and, if it is there, cyberpunk_convai_ambient_driver_guard.archive

To remove everything, also delete %LOCALAPPDATA%\AiNpcCyberpunkMod (the memories stored on your computer, saved keys, logs, reports and any models you added in its local-ai folder).

Free and paid#

Free and paid. The mod is free, but some AI services charge. Free: Local AI, and a free API key from Google Gemini (no card), OpenRouter (its free models) or Groq (its free tier). The Pocket and OmniVoice voices run on your PC for free, and so do the game's own dub voices. Paid: the OpenAI API needs prepaid credit and is separate from a ChatGPT subscription, so ChatGPT does not pay for it. The Anthropic API is paid too, and a new Mistral account may need a paid plan, which also covers the Voxtral voice. When a Custom account has no credit left, the game says so once per session, names the service and lists the free choices. It never switches AI by itself.

Already pay for OpenRouter or Nano-GPT, or run Ollama or LM Studio on your PC (or on another PC: Run the AI on another PC)? The Custom (OpenAI-compatible) provider uses any of them: paste an address, a model ID and, for most, a key (details).

What you can do#

• Talk by voice or text. Face someone, press your talk key, and speak or type. They answer out loud with subtitles, a moving mouth, and a face that changes at the very word where the feeling shifts. Facing nobody, the key talks to the first member of your party, so you can chat while you drive without turning to look.
• Bring a crew. Invite companions from the Companion Manager. They walk beside you, run to catch up, follow you through doors, lifts and fast travel, sit with you on couches and bar stools, settle into seats and standing spots when you stay in one of your apartments (and get up and follow when you leave), ride in your car and fight with the game's own combat AI. Ask one to drive and they take you to your waypoint (with no car near, they call yours first). The Model browser lists every character model in the game that can work as a follower (photo mode puppets, which have no AI, are left out), and any of them can tag along.
• They have vision. With Companion vision on (it is off by default), each line you send carries a fresh glimpse of the game window, so a companion can talk about the place you are actually standing in. It works with every conversation provider.
• The whole street talks. Gangers, NCPD officers, guards, ripperdocs, vendors, bartenders and passers-by answer as their kind of person. Shopkeepers you keep coming back to remember you. Children in the crowd talk and sound like children, with a boy voice and three girl voices.
• The story cast answers too. Talk to the game's own Judy, Jackie, Panam or Rogue where they stand, whatever the game is doing with them (a scene, a fight, a ride): their own voice, a face that follows the words, and no "busy right now". There is no setting to turn on.
• Johnny in your head. Turn it on in Settings and, once the story has put him there, give a key to Talk to Johnny by voice or Talk to Johnny by text (Companion Manager > Key bindings, no key is set for Johnny from the start) and Johnny answers wherever you look. He answers as the voice in V's head.
• Group conversations. Address two or three companions at once (up to six, with Group chat size) and they take turns, each one answering the one who spoke before.
• Phone Chat and holocalls. A Chat tab sits on V's phone beside Messages and Contacts. Text any character that has a profile here, wherever they are and not only the game's contacts, or call them and they pick up on a holocall, live on screen for the contacts the game itself stages video calls for.
• V speaks too. Lines you type are voiced in the V voice that matches yours, then the companion answers. You pick it in Voice AI and microphone > V's voice: Off, Typed lines or Typed and spoken lines (the default). The row under it, V's voice engine, picks the voice: Pocket (the default), Voxtral, or OmniVoice (only while your companions use OmniVoice, it adds a short wait before replies to typed lines). What you say by microphone is repeated while V's voice is on Typed and spoken lines, the default: V says it out loud like a typed line, and the reply follows V's line. Choose Typed lines to keep V quiet for what you say.
• They remember. Conversations stay with a character across saves. You can make one character forget, or all of them.
• Story quests (beta, off by default). With ArchiveXL installed, turn on Companion Manager > Companions and Night City > Story quests (beta), and a character you are close to can bring their own quest in conversation, tracked in the Journal with map markers and a route. They also text or call you first now and then (it is part of this setting, Characters text first has no row of its own). The quest system is very experimental and has not been tested thoroughly yet. Story quests turn Character bonds on with them.
• Relationships that follow your save. Judy, Panam, River, Kerry and Meredith know where things stand with V.
• Make them yours. Edit a character's bio, speaking style and memories in your browser, or make a character of your own (a name, a bio, a voice clip) for someone who has no profile (with Google Gemini, OpenRouter, Local AI or Custom). Give any story character a voice clip of your own, and the local voices copy it. New to the editor? Read the character editor guide.
• Seventeen interface languages, and characters that answer in the language you write in, with no setting to switch on.
• An optional Mature setting for adults. It is off by default and works with Local AI and Custom. See Mature content.

Only characters the mod spawns join your party. Story characters you meet in the world, such as the game's own Judy or Jackie, can be talked to where they stand, always: they answer in their own voice and their face follows the words. While the game has them in a scene the mod does nothing else to them, so the story plays out as written, and a request in your own words (come with me, for example) is judged on its own.

Requirements#

Developed and tested with Cyber Engine Tweaks 1.37.1, RED4ext 1.30.0, redscript 0.5.31, Codeware 1.20.5, Mod Configuration Menu 0.10.5 and Native Settings UI 1.96. Newer versions may work. There is no load order to set: the mod's files live in folders of their own, and the mod finds Mod Configuration Menu and Native Settings UI whichever of the two Cyber Engine Tweaks loads first.

Audioware and TweakXL are no longer needed. Voices play through the mod's own audio, so game patches that break Audioware no longer silence the mod. If you still have either installed, nothing changes for you.

First run#

1. Check your keys. Five actions come with keys: F7 talk by voice, F6 talk by text, F10 group voice chat, F8 group text chat and F4 the Companion Manager (see Controls). Change any of them in Companion Manager > Key bindings (press an action's Key or button row, then the key or controller button you want) or in Cyber Engine Tweaks' Bindings. A key you had set already stays yours: the mod only fills actions that had none, and never overwrites one.
2. Say something. Face a character, press talk by text, type, press Enter (Esc closes the box and keeps nothing). Click outside the box instead and it closes at once but keeps your words, so you can walk and look around, and talk by text opens it again with them. Facing nobody, the key talks to the first member of your party, the one at the top of Companion Manager > Current party (for a group chat, that member and two others from your party picked at random). Or hold talk by voice while you speak and let go: it records from the moment the key goes down, so you can start talking at once, and letting go sends your words. (Voice AI and microphone > Voice activation can do it with no key at all, and on a controller you can give talk by voice a button such as the left stick click on the Key bindings page.)
3. Invite a companion. Open the Companion Manager (your key, or Esc, then Mods, then AI Companions (LLM); with Mod Configuration Menu installed the settings are in its list, under AI Companions (LLM)), choose Characters to invite and press Invite. To find someone fast, press Search by name at the top of the list (or, in Mod Configuration Menu, click its Search box at the top right), type part of a name and press Enter: the list shows every match on every page, and Clear search shows everyone again. When the "joined" notice appears, go back to the game and she is there.
4. Make it yours. In the Companion Manager choose Settings. AI provider and Voice AI and microphone are the two pages that change the most, and the other pages are Conversations, Companions, Night City, Interface and Privacy and help. There is no Apply button: a change is saved a moment after you make it, and a row that cannot be used goes back with a line under Back that says why.

Out of the box the voice is Pocket TTS, each character sounding like themselves, running on your processor. For the replies you choose a provider first: nothing answers until you do, and the game tells you once how (a free Google Gemini key, or Local AI).

Controls#

You can also set these seven actions inside the game: Companion Manager > Key bindings. Press an action's Key or button row, then the keyboard key or controller button you want (Esc or B cancels). It works at once, Clear removes it, and the Cyber Engine Tweaks bindings above keep working next to it (a key set in both runs the action once). A key can be a combination: hold Shift, Ctrl or Alt while you press it, for example Shift+F5. A combination runs only with exactly those keys held, and a plain key does not run while one of them is held (so F7 and Shift+F7 can be two different actions). The game uses Shift to sprint, Ctrl to dodge and Alt for the weapon wheel, so V does that as well while you hold one. The five default keys come back with Restore defaults. A controller button works the same way (for example the left stick click for talk by voice, which replaces the old Controller talk button).

The microphone only ever starts when you press the key. A recording stops by itself after 30 seconds, and opening a menu cancels it. If you see V keep walking after typing a message, tap that direction key (W, A, S or D) once.

Pick a brain: conversation providers#

Companion Manager > AI provider chooses who writes your characters' replies. A change takes effect when no conversation is running.

Measured on one laptop; yours will differ, and Google's allowance is Google's to change. The Custom column is whatever you point it at, so it has no figures of its own.

Local AI#

The language model runs on your own graphics card. No account, no key, no internet. Nothing else to set up: extract the pack into the game folder like the mod itself, and the mod starts the AI, loads the model and stops it again by itself. No server, no Ollama or LM Studio, no command line.

1. Download a Local AI pack from the mod's Nexus Files tab and extract it into the game folder (Install).
2. Start the game. The Local AI model row shows "(Warming up)" while your card prepares the model, about a minute the first time.
3. Choose Local AI as the provider.

Install neither, one or both. The Local AI model row lists what is installed, smallest first; until you pick one, the largest that fits your free video memory is used. You need a Vulkan graphics card (NVIDIA or AMD), at least 16 GB of system memory (32 GB is better), and about 5 GB more than usual while it runs. Integrated graphics alone is not enough.

• The model unloads after ten quiet minutes so the game gets its video memory back, and reloads in a few seconds when you talk again.
• Short on memory, the mod steps down by itself and keeps your voice before the model: Local AI Ultra gives way to Local AI, and if that does not fit either, you get one notice and Local AI tries again on your next turn.
• You can add a GGUF chat model of your own, or use a server you already run (KoboldCpp, LM Studio, Ollama or any other OpenAI-compatible one; no pack needed, and it shows as "(custom)" in the Local AI model row). Keep the model and its manifest.custom.json in %LOCALAPPDATA%\AiNpcCyberpunkMod\local-ai and they survive reinstalls and mod managers; the Local AI folder inside the mod works as it always did, and both are read.

Google Gemini (API key)#

1. Make a free key at Google AI Studio. A Google account is enough, no card. Keys start with AQ. or AIza.
2. Copy it, open AI provider and press Add on the Google Gemini API key row. The address and the model are fixed, so that is the only thing to paste.
3. Choose Google Gemini (API key) as the provider.

The free cloud option: fast replies, about a second to the first words, and light on your PC, because nothing runs here. Add checks the key once, and the key row says whether Google accepted it. Remove deletes it. The replies are plainer: a small fast model holds a character less tightly. Memory stays on your computer, the microphone is heard by Google Gemini with the same key (choose This PC under Speech recognition to keep it on your computer, English, with other languages from a Microphone pack), and companions have vision of where you are when Companion vision is on. The voices are Pocket, OmniVoice or Voxtral. With Google Gemini chosen but no key saved, nothing answers and you see one notice that says so. Mature content never goes to Gemini. It also writes for the characters you make and for add-on speakers, like the Custom provider.

The key goes to Google's address and nowhere else. When the free limit is reached you see the usual limit notice: wait for the reset, or switch the provider.

OpenRouter#

1. Make a key at openrouter.ai/keys. A free account is enough, no card.
2. Copy it, open AI provider and press Add on the OpenRouter API key row. The address and the model are fixed, so that is the only thing to paste.
3. Choose OpenRouter (your key) as the provider.

The mod uses a free model that can see pictures (google/gemma-4-31b-it:free), so a key with no credit works at once. Add checks the key once, and the key row says whether OpenRouter accepted it. Remove deletes it. OpenRouter's limit for free models is 20 requests a minute and about 50 requests a day (roughly 20 conversation turns), or 1,000 a day once you have bought 10 dollars of credit, which free models never spend. Memory stays on your computer, the microphone is recognised on your computer, and companions have vision of where you are. The key goes to OpenRouter's address and nowhere else. Mature content is not available with it (use Local AI or Custom). To use a different model, open the Custom endpoint page and use the OpenRouter steps there.

Mistral#

Mistral used to give everyone a free API key with about 5,000 turns a day, but new accounts don't seem to get one any more (October 2026), so a new key may need a paid plan. It left the provider list on 30 October 2026 and is back in it as a main provider since 1 November 2026: the third entry, Mistral (your key), after Local AI and before OpenRouter.

1. Create an API key at Mistral.
2. Copy it, open AI provider and press Add on the Mistral API key row. It reads the key from the clipboard, checks it with Mistral and saves it on your computer.
3. Choose Mistral (your key) as the provider.

Fast, with plainer replies. It is one key with two uses: it also powers the Voxtral voice (Pick a voice), so the Voxtral key row of Voice AI and microphone shows the same key, and a key added there is the Mistral key here. With both, the whole conversation runs in the cloud. Remove on either row deletes the key and the help text says what stops working. Mistral is offered here only: the Custom endpoint page has no Mistral preset any more (2026-11-03). A player who had set up Custom with it keeps Custom as the provider, with replies working as before, and the key of that setup is saved as the Mistral key when none is saved. A saved Mistral choice keeps working too, with nothing to migrate.

Custom (OpenAI-compatible)#

Any endpoint that speaks the OpenAI chat-completions protocol: OpenRouter, Nano-GPT, or a server on your own computer such as Ollama or LM Studio. It is the sixth entry of the provider selector and the last in the list.

1. Open AI provider and press Open on the Custom endpoint row, under the Google Gemini key row. Its page holds the rows below. Back to settings returns.
2. Copy the endpoint's base address and press Paste endpoint URL: https://openrouter.ai/api/v1 for OpenRouter, https://nano-gpt.com/api/v1 for Nano-GPT, https://api.groq.com/openai/v1 for Groq (a free tier), http://localhost:11434/v1 for Ollama, http://localhost:1234/v1 for LM Studio.
3. Copy your API key and press Paste API key. A server on your own computer needs none.
4. Copy a model ID from the provider's list and press Paste model ID, for example meta-llama/llama-3.1-8b-instruct on OpenRouter or llama3.2 on Ollama.
5. Press Test connection: it sends one tiny request and says OK, or what the provider answered, in plain words.
6. Go back and choose Custom (OpenAI-compatible) as the provider.

Each paste reads the clipboard once, checks it and saves it on your computer, and Remove deletes it. You can also write the three values into player-custom-endpoint.txt in %LOCALAPPDATA%\AiNpcCyberpunkMod\data. Only the last four characters of a key are ever shown, and the key goes to that endpoint and nowhere else. An http:// address works only for your own computer or home network, and any other host needs https://. Nothing is sent until you choose the provider or press Test connection. Without an address and a model ID, nothing can answer and one notice says so. When the provider answers with an error you see a notice that says what to do: key not accepted, no credit (it names the service and lists the free choices), unknown model ID, rate limited, or no answer. Memory stays on your computer, the microphone is recognised on your computer in English (or by a Gemini key, or by a Custom speech endpoint, below), and the voices are the same as with Google Gemini (the Voxtral voice needs a Mistral key). Mature content stays with Local AI. The page can also serve embeddings (memory search) and speech recognition (your microphone), each behind its own switch (Use for embeddings, Use for speech recognition, both off) with its own address, key and model ID and its own Test connection. They speak each provider's own format: Gemini, Cohere, Mistral, NVIDIA, Deepgram, AssemblyAI, ElevenLabs and Groq, as well as the OpenAI form. To hear you through it, choose Custom endpoint under Speech recognition in Voice AI and microphone.

Run the AI on another PC#

This guide has a page of its own: Run the AI on another PC.

Mature content#

AI provider > Mature content is off by default and for adults only. It needs Local AI or Custom, never Google Gemini or OpenRouter. Turning it on beside Google Gemini or OpenRouter picks the first of these that you have: Local AI (when its pack is installed), then Custom (when it is set up). With neither it stays off. Choosing Google Gemini or OpenRouter turns it off again. Voxtral voices are not used with it: a companion voice or V on Voxtral moves to Pocket, and comes back when you turn it off. The Mature packs below are for Local AI. With the standard models it changes language and tone, because those models refuse most adult material. The two optional Mature packs, Mature and Mature Ultra, on the mod's Nexus Files tab, add one model each without those refusals. The Local AI engine and the memory model are both in the Runtime, so either works on its own, without a Local AI pack (with a Mature pack alone, Local AI runs only while Mature content is on). Mature is about 2.9 GB to download and needs about 3.4 GB of free video memory, Mature Ultra is about 5.5 GB and needs about 6.5 GB. Each has a vision file in the vision folder of mature (0.68 GB for Mature, 0.56 GB for Mature Ultra), used only while Companion vision is on. Fixed limits stay on for every model, with or without the packs. Details are in the notes inside the pack.

Pick a voice#

Companion Manager > Voice AI and microphone > Companion voices. Switch any time. A voice that cannot run is replaced by a simpler one and the game tells you once.

Balanced and Quality use 16 generation steps, High Quality and Ultra Quality 32. Balanced and High Quality copy a short sample of the original voice, Quality and Ultra Quality a long one.

• Pocket: nothing to do.
• OmniVoice: install the pack, restart the game, pick an OmniVoice entry. Without the pack the entries stay listed, the row reads "(no Omni pack)" and choosing one is refused with a line under Back that says it needs the pack. The very first start takes two to three minutes (longer on a laptop in Silent power mode), later starts are faster. Until OmniVoice is ready, or when a start fails, replies show as text, and the mod retries in the background. It never puts Pocket in its place. AMD cards (experimental): on a PC with an AMD Radeon RX 6000, 7000 or 9000 card and no NVIDIA card, Voice AI and microphone has an OmniVoice for AMD page: one button downloads AMD's own ROCm PyTorch (about 1.5 GB, 4 GB on the drive), tests it on the card and OmniVoice then starts on it. Your own Python (omni-python.txt) still works and wins.
• Voxtral: save a Mistral key (press Add on the Voxtral key row right under Companion voices, or on the Mistral API key row of the AI provider page, Mistral), then pick Voxtral · Mistral cloud. A Mistral key you already saved on the Mistral API key row shows there as set with nothing to paste again. Until then it reads "Voxtral · needs Mistral key".

If the local voice you chose cannot run (OmniVoice or Pocket), your companions' replies show as text with one notice, and the voice comes back by itself when it recovers. The mod never moves you to a cloud voice behind your back. Voxtral has no fallback either, local or cloud: a line it cannot voice (no key, a cool-down, a language it does not speak) is shown as text, with one notice, and it speaks again by itself when it can.

The same page (Voice AI and microphone) also holds Speech recognition (This PC, Google Gemini, Voxtral or Custom endpoint: Automatic, the default until you choose, uses Google Gemini once a Gemini key is saved, else This PC), Voice language (English by default, Automatic, which with Mistral lets you mix languages by voice because Voxtral hears the microphone with no language set, the game's other languages, then Finnish, Swedish, Norwegian, Danish, Dutch, Greek, Romanian, Bulgarian, Croatian, Serbian, Slovak, Slovenian, Estonian, Latvian, Lithuanian, Hindi, Indonesian, Vietnamese or Hebrew in the same list: those need a text provider and a voice that speaks them, Voxtral or OmniVoice, never Pocket), Companion voice volume (a slider from 50 to 300 percent, 150 by default, for the companions' and characters' voices), V's voice (Off, Typed lines or Typed and spoken lines, the default), V's voice engine, V's voice volume (its own slider, 50 to 300 percent, 150 by default, for V's spoken lines only, and the game's own audio is untouched), Natural speech (pauses, hesitations and, with OmniVoice, laughs and sighs, off by default), Microphone, Voice activation, Speak dialogue choices and Lip sync delay (Bluetooth). Subtitles (Off, Cinematic subtitles or Above the speaker) and Subtitle language (Same as speech by default, or any listed language, translated by one extra short request to your own AI provider while the voice plays) are on the Interface page. The engine is one list: Pocket · On your PC (the default, it runs on your processor and starts in about 0.2 s), Voxtral · Mistral cloud (made in the cloud with your Mistral key, loads nothing on your PC, speaks English and many other languages, and greyed out until you save a key) or OmniVoice. It is one engine for typed and spoken lines, and it is separate from the companions' voice, so any mix works: Pocket for V beside Voxtral companions, or Voxtral for V beside Pocket companions. Choosing an engine does not turn V's voice on or off. V's voice is prepared before your first line, so it is normally audible well before the reply. If it is still not ready when the reply is, the reply waits for V's first audio (about 30 s at most without audio) and then follows V's line, and V's words always show as a subtitle.

Smooth frames, fast replies#

The heavy lifting happens in a separate background service, not inside the game, so cloud text (Google Gemini, for example) with the Pocket voice uses no video memory at all. Pair Google Gemini with the Voxtral voice and the companions' voices come from the cloud: no companion voice model on your processor or graphics card. V's typed lines follow V's own voice settings: Pocket (the default engine) runs a small voice model on your processor, so set V's voice to Off, or V's voice engine to Voxtral, for nothing local at all. Only OmniVoice and Local AI touch your graphics card, and the smaller Local AI model keeps fully local play light.

Every voice sounds like the character, Pocket and Voxtral included, so pick by what matters to you:

• Least effect on your frames: go all cloud. Google Gemini for the text, Voxtral for the speech. Nothing runs on your PC for the companions (set V's voice to Off, or V's voice engine to Voxtral, to keep V's line off your processor too). Cloud text with Pocket is nearly as light: no video memory, a little processor time while it speaks.
• Fastest replies: Google Gemini (or Mistral) for the text with Pocket for the voice. Pocket starts speaking in 0.15 to 0.2 s, and a reply began speaking a second or two after sending. Voxtral starts in about a second.
• Fully local, best replies: Local AI Ultra, with Pocket for the voice (or OmniVoice on an NVIDIA graphics card). No account, no internet.
• Fully local, lightweight: Local AI with Pocket, about 1.8 GB of extra video memory.
• Extra expression: OmniVoice can laugh and sigh, and Natural speech adds pauses and hesitations. OmniVoice is the heaviest voice and starts speaking a little later.

A few habits help every setup:

• Set Windows power mode to Balanced or Performance, and plug a laptop in. Silent mode cut Local AI's speed by about 40 percent in testing, slows speech recognition and stretches OmniVoice's first start to about three minutes. On a laptop it is the biggest single factor.
• Text only? Set Companion voices to Off and the voice models are released.
• Unused models leave on their own: Local AI after ten idle minutes, OmniVoice when playback is off, the microphone's recogniser 90 seconds after your last recording.

Extra video memory each setup uses on top of the game's own:

Settings#

Everything lives on the Companion Manager Home page, a flat list: Main characters to invite, Side characters to invite, Parties, Relationships, Character editor, AI provider, Voice AI and microphone, Conversations, Companions and Night City, Content packs (only when an add-on or a content pack is installed), Model browser, Key bindings and Privacy and help. Open a page, and Back to Companion Manager returns to Home. Some rows open a page of their own and go back to the page they came from. Restore defaults, on a page, puts back the settings of that page only. Defaults are for a new install; your saved choices always win.

A few worth knowing:

• Talk to ordinary NPCs (always on, there is no switch): passers-by, vendors and the rest answer as their kind of person. They do not know your story and forget you after a save load, except shopkeepers you come back to.
  • A talk can have consequences. If you clearly threaten or seriously insult an ordinary NPC (not on a joke or rude banter), they end the talk and walk or run off, call the police (V gets one wanted star), or, for a ganger, merc or guard, turn hostile and attack. The character decides in its own reply, and only ordinary NPCs are ever affected: never story or quest characters, never in a scene, a fight or a safe zone.
• Street reputation (always on, no setting): the people of Night City remember V. Each gang, the NCPD, the corporations and the locals of each district have their own standing with V, built slowly from your Street Cred, your choices in jobs (who you paid, spared or crossed), fixer gigs done in their district, who you fought (only fights they saw count, a quiet takedown is just a rumour a day later), crimes the police saw, and how you treated people in talk. It fades over a few in-game weeks. It only changes how ordinary people greet you, trust you and help you: a Maelstrom ganger who has heard you shot Royce is cold and may tell you to get lost, the locals of a district you helped are warmer. Named and story characters are never affected, and nothing in the game itself changes. It works from events, with no work every frame.
• Talk to story characters (always on, no setting since 4 Nov 2026): the game's own named characters (Judy, Jackie, Panam, Rogue and the rest of the cast) answer where they stand, in their own voice, with their own memory and a face that follows what they say. Nothing the game is doing with them stops a talk: a scene, a quest conversation, a commanded AI, a fight, an injury, a ride, a call or a cutscene never shows a "busy right now" notice, typed, by voice, by Phone Chat or in a group. Only a character that is no longer there ends a talk. A reply is never held or dropped for a scene. While the game has the character in a scene the talk is dialogue and face only: the mod does not make them follow, sit or gesture, and the scene keeps their face. The probe log keeps one line per talk, "would have refused before", with the reason the old rules gave, so a report still shows where the game had the character. Old saved values of the former Talk to story characters list are ignored.
• Talk during missions (experimental, part of Talk to story characters, always on): a game-placed story character you talk to during a story mission joins in: talk in every form (typed, voice, Phone Chat, group) works in combat, on foot and in a vehicle, and they get combat remarks, speaking up, their idle face, talk gestures, and follow you on foot and ride in your car where the game lets them. The game always has priority: whenever a mission script drives the character (a scripted move, a cutscene, a scene, a dialogue choice, a scripted ride) the mod steps back and resumes afterwards, and a talk then goes on as dialogue and face only. It never teleports or despawns them or touches quest or story state.
• Johnny in your head: once the story has put Johnny Silverhand in V's head (after Playing for Time) and until an ending takes him out, the Talk to Johnny by text and Talk to Johnny by voice keys reach him wherever you look (the talk keys with nobody in front of you now go to your party, and the group keys never include him). He answers as a voice in V's head, with his own voice and memory, knows your jobs and your story so far, and will tell you which job he would do next. He stays quiet during scenes, calls, dialogue choices and fights, and when the game gives him a line. One list: Off (the default), On (he answers when you talk to him and says nothing unprompted), Speaks up rarely, Speaks up normally or Speaks up often (see the next point).
• Speaks up rarely / normally / often (the last three choices of Johnny in your head): he makes a short remark of his own after you meet a named character in the game's own dialogue, finish a main or side job, a romance moves on, you arrive somewhere he knows (the Afterlife, Pacifica, the Badlands, Dogtown, Corpo Plaza, Viktor's clinic and a few more, each at most once every few hours), the Relic acts up, or a big fight ends (counted from the game's own tallies, nothing is hooked into the fight): about 2 an hour when rarely, about 4 when normally, about 6 when often (random gaps, counted in calm play only), never during scenes, calls, fights or while you talk. He may also follow up on a companion's own comment or on a reply in a conversation you are having, and then he is prepared while she is still talking. With companions speaking up too, the two rates add up and they take turns, at least 2.5 minutes apart.
• Companions speak up (Conversations, Off, Speaks up rarely, Speaks up normally, Speaks up often, Speaks up very often, with Speaks up often as the default (a choice you made stays, and a file written by the 2026-11-04 build that never chose it moves once to Speaks up often)): your party companions sometimes start talking on their own, now and then at a calm moment, and after a scene or conversation they were part of (with the mission and what was said in mind). When two or more party companions are near, a calm speak-up is most of the time a short chat among up to three of them instead (3 to 5 lines, one voice at a time, then quiet until the next one), and it stops the moment you talk. About 2, 4, 6 or 12 an hour of calm play (Speaks up very often also speaks while you ride in a car, when you arrive somewhere new and when the weather or time of day changes), never during fights, scenes, calls, menus or while you talk. Lines are at least 150 seconds apart, so they never push a free Google Gemini key towards its per minute limit.
• Immersive invitations: Off by default. On, inviting a companion (the Invite button, the Contacts tab, or asking her in a text) is an appointment: she appears out of sight, walks or drives in, texts you when she is there, waits, and joins you with a greeting when you reach her, or leaves angry and remembers it. Choose how she arrives (Walk or Car), when (right now, or in 2, 5 or 10 real minutes of play) and how long she waits. She comes to where you are; ask for your map marker to send her there.
• Housemates: Off by default. On, a character you tell she can stay at your place (in person, by text or on a call) moves into one of your apartments and lives there: she sits, sleeps and goes about her day, greets you when you come home, and you can call her along. Housemates also talk to each other (one list, Housemates chat: Off, Now and then, Often, Often by default), you can choose which apartment a character moves into when you do not name one, and one button moves every housemate out at once. Off, nothing of this runs and nobody offers it; turning it off moves them out of your apartments but remembers who lived there, so turning it on brings them back.
• Revenge ambushes (new, off by default, Companions and Night City): a gang you keep killing may, rarely, send a small squad of its own fighters after you. They start out of sight, run at you, and one of them shouts a short threat in its own words (an AI line in the gang's voice). It counts the gang members you kill the way Street reputation does, waits a long time between squads, and never comes in a fight, a scene, a call, a vehicle, indoors, in a safe zone or while you are wanted. The game has no system you can call for this (its gang chases are scripted into quests), so the squad is the mod's own, built from the game's own retaliation fighters.
• Companions change outfits (new, off by default, Companions and Night City): party companions change into other looks of their own by the game's time of day, now and then for variety, and into casual clothes at home, only while nobody is looking and only for the mod's own party copies. Fifteen characters have extra looks, everyone else keeps theirs.
• Romance animations (always on, no setting): with a character you are in a romance with, the character chooses a kiss in her own reply, the way she chooses her other actions, and the two of you play the game's own two person animations, with a screen blink in and out. Asking in a talk ("kiss me") works too. There is no menu and no question. Never in a fight, in a car, in a scene, with a menu open or during a mission.
• Replies that read the moment: strangers no longer talk as if they knew V, "I gave you eddies" is read as V paying the character, and a companion knows which weapon they carry, also holstered (the last one follows the Company and vehicle switch). Custom characters spawned from another mod's menu are recognised by their model record, so they keep their profile.
• Dynamic info: which live game info the characters are told. Twelve switches, all on by default except Other mods’ state: Location, Time and weather, Company and vehicle, Health, heat and fights, V's money, V's gear and clothes, V's gender and lifepath, Quest and journal, Time since last talk, Music and TV, Other mods’ state and Recent game dialogue. An off kind is left out of every conversation provider (Google Gemini, Mistral, Local AI and Custom) and the mod stops reading it from the game. A character mentions where you are, the time or the weather only when it matters to what you say; with Location off nobody is told where you are at all. What each character knows about your story stays on. The page has the same switches twice: In person (your party and face to face talk) and Phone calls and texts (holocalls and Phone Chat), so a companion with you can be told things a character you only call or text is not. Both start the same.
• Memory: Forget selected character and Forget all memories (the row reads "Forget all memories (6 characters)" and its button "Forget"), each with a confirmation. A forget also clears the matching Phone Chat messages. Relationships has Auto, On or Off for Judy, Panam, River, Kerry and Meredith; Auto follows your save.
• Your language is automatic: there is no setting for it (the old Multilingual conversations switch was removed). The mod detects the language of each message you type or speak, and the character answers in it. On Google Gemini, OpenRouter, Local AI and a Custom endpoint the mod tells the model which language your message is in, and says nothing when you play in English. The voice is never moved or refused for it. It speaks the reply with OmniVoice, or with Voxtral for the languages Mistral lists for it (more than twenty, since Voxtral TTS 3: Mistral's cloud voice speaks the language the reply is written in, in the character's own cloned voice); with Pocket, which speaks English (and French, Italian, German, Spanish and Portuguese when the optional PocketTTS model and the multilingual pack of that language are installed), or in a language the chosen voice does not speak, the reply shows as text, and the mod says so once per session. No voice model is loaded for it. Your language is remembered between launches, so the microphone starts in it. With Mistral chosen as the provider (kept from before) the microphone understands other languages too (Mistral transcribes the recording with your own key); with Google Gemini, OpenRouter, Local AI or a Custom endpoint the optional Microphone packs do the same on your computer.
• Eye contact and head tracking: companions' eyes and head follow you, with nothing to switch. Separately, and with nothing to switch, they gesture through longer replies, nod while you talk and hold idle poses beside you; gestures stop the moment anyone moves or a fight starts.
• Subtitles (Voice AI and microphone): Off shows none. Cinematic subtitles use the game's dialogue line, Above the speaker floats them over the character.
• Companions react (Companions and Night City): a nearby companion makes one short comment when a fight starts and ends, or after a run of loot pickups, at most once every ten minutes each. It is one switch, on by default again from 2026-11-05 (it was off for new installs on 2026-11-04 only, and a choice you made stays). A battle is an opening and a closing line per ten minutes, loot is one comment per ten minutes, or per two minutes for newly acquired Legendary gear.
• Invulnerable companions (Companions and Night City, off by default): the companions in your party take no damage and cannot die, which helps with mods that make the game much harder. It covers the mod's own party copies only, never characters the game placed.
• Send error reports automatically: when something in the mod goes wrong, a short report goes to the developer so it can be fixed without asking you for logs. See Error reports below for what it holds and what it never holds.
• Developer mode is for testing and bug reports: routine popups, diagnostic traces and developer-only experiments. You will not need it to play. While it is on, the service also writes each turn (who spoke, the provider, your words, the reply, whether vision went along and where you were) to dev-transcript.log in %LOCALAPPDATA%\AiNpcCyberpunkMod\data, about 2 MB at most, with no keys in it. It is never sent anywhere or added to a report, and nothing is written while the setting is off.

Microphone packs#

English speech is recognised on your computer by a small model that is already in the Runtime. For other languages:

• Mistral as the Conversation provider (kept from before): the recording is transcribed by Mistral with your own key (nothing to install).
• Google Gemini with a saved key: Gemini hears you in any language, no pack needed (Speech recognition > This PC keeps it on your computer and uses a pack instead).
• OpenRouter, Local AI or a Custom endpoint: your speech never leaves your computer, so install a pack. European languages (Russian, Ukrainian, Polish, Czech, Hungarian, German, French, Spanish, Italian, Portuguese, about 0.67 GB) or Asian languages (Mandarin Chinese, Japanese, Korean, about 0.24 GB). Extract the zip into the Cyberpunk 2077 folder like the other packs. Nothing needs switching on: the mod uses the pack for the language you have been typing or speaking (remembered between launches, so a French player with an English game is heard in French from the first recording), else the game's language.
• A pack runs on the processor, one thread at high priority (the mod gives everything on the way from your words to the reply's first sound high priority), and is loaded only while you speak in its language; it leaves memory about a minute after your last recording. Measured on the development laptop: about 4 to 5 seconds to load (the first load after installing can take a minute while Windows checks the file), then about a quarter of a second of processing per second of speech, so a 5 second sentence is read in about 1.5 seconds after you press the key. Words appear after the second key press, not while you speak.
• Without the pack for your language you see one notice when the recording opens, and the recording is read as English.

Troubleshooting#

A character does not answer. Give the game a moment after launch: the background service starts with it. Privacy and help > Help and support shows the service status, if it is not running, restart the game. With Google Gemini, check that the key row says it was accepted. With Local AI, check that the Local AI model row does not read "(Warming up)", "(Not enough memory)" or "(Unavailable)".

The game says no provider is chosen, or that the shared free tier has ended. The shared free tier is gone, so on a new install, or one that used it, nothing answers until you choose a provider. On AI provider, choose Google Gemini (API key) and add a free key from Google AI Studio, or choose Local AI (with a Local AI pack).

V's lines are never spoken. Lines you type are voiced by default, in Voice AI and microphone > V's voice (Typed and spoken lines, with the engine Pocket unless you changed it). Lines you speak by microphone are repeated too unless you set V's voice to Typed lines. If you type a line, V's voice is not Off and V is still silent, open Privacy and help > Help and support and press Save report, so the cause shows.

The microphone does nothing, or says it needs speech recognition. Check that talk by voice has a key in CET's Bindings. Hold it while you speak, then let go. Windows must let desktop apps use the microphone, and yours should be the Windows default input device. With Mistral or Local AI your speech is recognised on your computer, in English, and the recogniser takes a moment to load the first time: if it is not ready yet you see "Speaking to companions needs speech recognition, which this provider does not have yet. Type your message instead." Try again in a moment, or type.

Voice input says it only understands English. You spoke another language and nothing here can read it. With Mistral kept from before, check that the key row says it was accepted. With OpenRouter, Local AI or a Custom endpoint (or Google Gemini with This PC chosen), install a Microphone pack for your language (the notice names it when one exists). Typing always works in any language.

OmniVoice takes two or three minutes the first time. That is normal for a model this size. Use Balanced power mode, not Silent. Replies show as text meanwhile, a held notice reads "Preparing ...", and later starts are faster.

Replies show as text when I picked OmniVoice. Without the pack choosing an OmniVoice entry is refused (the row goes back to the saved voice and says it needs the pack). With it, an OmniVoice that fails to start leaves the replies as text, with one notice, and retries in the background; choosing it again also retries. Nothing is sent to a cloud voice meanwhile.

Only some characters have an OmniVoice voice. The pack holds most of the characters' reference clips (the full install carries only a few), so a pack that stopped extracting half way, or a folder a mod manager cleared, leaves the others without a voice and their replies are written. The bridge log then has one [OmniPack] ... reference clips are missing line per clip length, naming a few of them. Extract the OmniVoice pack again over the game folder (replace files if asked): it only adds what is missing and never changes the full install's files.

Lips run ahead of the sound. Bluetooth audio arrives a little late. The lip sync delay is automatic: it adds 150 ms while Windows plays through Bluetooth. Hearing nothing at all? Check the game's Master and Dialogue Volume and the Windows mixer.

A companion cannot be summoned there. Companions need clear space, and a spawn is refused when there is no safe spot nearby. Try somewhere more open. Party membership is not restored after a save reload, a game restart or a Cyber Engine Tweaks reload.

Does it change the story? No. Companions are temporary copies that the mod owns. They never recruit, revive or alter story characters. A story character in the world answers when you talk to it, with its voice and face, and the mod otherwise stays out of scenes, quest conversations, fights and rides.

Which of my data leaves my computer? With Google Gemini your lines go to Google, along with a recent image of the game window when Companion vision is on, taken only while someone you can talk to is near, kept in memory for a few seconds and never saved. With Mistral they go to Mistral. With a Custom endpoint they go to the endpoint you chose, and only after you choose it (a server on your own computer keeps them there). With Local AI nothing leaves your computer for replies, unless you point it at a model server on another computer. With a Gemini key, the recording of what you say also goes to Google. Voxtral sends the text to be spoken and a short reference clip of the character to Mistral. Error reports, below, go to the mod's own server.

Error reports. When something in the mod goes wrong, it sends a short error report to the developer's server, so problems can be fixed without asking you for logs. It is on by default, Privacy and help > Send error reports automatically turns it off, and then nothing is recorded or sent. The first time you load the game after updating, a notice says so once.

• What a report holds: the error and where in the code it happened; the versions of the mod, the game and the required mods; your conversation provider, voice and Local AI model choices; whether Mature content is on (on or off, nothing more); what kind of PC it is: your graphics card's name and memory, the PC's total RAM and the memory limit Windows sets (RAM plus the paging file), the processor, the Windows edition, build and display language, the security products Windows lists (such as Defender) and whether Smart App Control is on, whether the game runs as administrator, the game's drive (kind, free space, whether it is under Program Files), and the screen's size and refresh rate; the names of your installed mods (the folder and file names of Cyber Engine Tweaks mods, RED4ext plugins, redscript mods and archives, never a path); the setup check's lines; and up to 150 log lines from around the error (for a setup problem or a script error these include the error, warning and failed-load lines of the Cyber Engine Tweaks log and scripting log, which can mention other mods), with keys, IDs and your Windows user name removed. Your installation is identified only by a random number hashed twice.
• What it never holds: API keys or tokens, what you or the characters say, typed lines, character memories, microphone audio, or your Windows user name. Your IP address is not stored.
• How often: reports are sent in the background, never while a character is answering. Reports are kept for 30 days. The support report shows whether the setting is on. Details: Error reports and privacy.

Replies are slow, or Local AI runs out of time. The model, the voice and the game share one graphics card. When the game fills it, the model runs mostly on the processor. Lower the game's texture quality or ray tracing so about 4 GB of video memory is free, or choose a smaller Local AI model. Google Gemini (a free key) and the Custom provider use no video memory.

Replies are written but never spoken. If your reply language is not English, Pocket needs the matching Pocket Voice language file. Install it from the Files tab (extract it into the Cyberpunk 2077 folder like the other packs), or set the reply language to English. With another voice engine, check Voice AI and microphone > V's voice and the voice engine row.

OmniVoice is slow, or replies show as text with it. OmniVoice needs several GB of free video memory next to the game. On a small card (8 GB or less) choose Pocket Voice (install its file) or a cloud voice, or lower the game's graphics settings. The first load takes two or three minutes, and replies show as text meanwhile.

Pocket shows the text first, or a stranger sounds different. On a PC with about 16 GB of memory, Pocket waits for free memory before it loads a character's voice. The reply shows as text at once and the voice follows on later lines. When memory is short, strangers get a stand-in voice. Close other programs, or enlarge the Windows paging file, and restart the game.

Gemini or Voxtral cannot be reached. Your network or country may block Google or Mistral. Use a VPN, or choose Local AI or the Custom provider (and a Pocket or OmniVoice voice instead of Voxtral).

My buttons do nothing in the settings. Buttons wait while changes are pending. Press Apply or Revert first, then press the button again.

Setup check says redscript or the compiler is missing, or the scripts did not load. Install redscript from the Requirements list for your game version and start the game once. If it is installed, open r6/logs/redscript_rCURRENT.log: when it names another mod, that mod stopped the scripts from compiling. A game version that does not match the Scripts or an old Codeware can also cause it.

Setup check says the install is incomplete, or Windows or my antivirus blocks files. Extract the main file again into the Cyberpunk 2077 folder (the one with bin, archive and r6), over the old files. Add the game folder to your antivirus exclusions, as it can quarantine the mod's files (Windows Security > Virus and threat protection > Manage settings > Exclusions), then restart the game.

Mod Organizer 2: Pocket Voice, a Pocket language, Speech to Text or the OmniVoice voices show as not installed. Before 0.5.3 the mod looked for these packs only inside the main file's own mod. Update the main file to 0.5.3. Each pack can then be its own mod or sit inside the main mod, both work. Tick each one in the left pane and start the game with Mod Organizer 2's Run button.

Mod Organizer 2: the mod or its packs are missing in game. Start the game with Mod Organizer 2's Run button. Started from Steam, GOG or a shortcut, the game sees none of the mods in Mod Organizer 2.

Mod Organizer 2 asks to Merge, Replace or Rename when I install a pack. Mod Organizer 2 may suggest the same name for every file of the Nexus page. Choose Rename and give the pack its own name, or Merge to put it into the main mod. Never choose Replace for a pack, as that removes the main file.

Mod Organizer 2 shows conflict signs between my packs. They are harmless. The packs share the same licence files and the HOW TO INSTALL note. Any order works.

Mod Organizer 2's Overwrite folder fills with files of this mod. That is normal. The mod keeps its settings and status files there (in bin\x64\plugins\cyber_engine_tweaks\mods\AiNpcCyberpunkMod\runtime) and they survive a restart. Leave them there, don't delete them or move them into a mod you might untick. Your memories, chats and keys are in %LOCALAPPDATA%\AiNpcCyberpunkMod, outside every mod.

The mod does not load, or the manager will not open? Post what you see (a screenshot helps) on the Nexus Posts tab or on Discord. If the game starts, Privacy and help > Help and support makes a report you can attach.

The game closes when I load a save. If the game closed during a load, the next start skips everything the mod restores (the last party, outfit changes, housemates, quest stages) and says so in one notice. Two closed loads in a row set the saved party and outfit files aside (renamed, not deleted). To clear it by hand with the game closed, delete party-restore.json, outfit-rotation.json, ambush-state.json, housemates.json and load-guard.json from the mod's runtime folder (with Mod Organizer 2 that is the same path inside the Overwrite folder) and from %LOCALAPPDATA%\AiNpcCyberpunkMod\data\game-state, and delete r6\cache\modded. The steps for Vortex and Mod Organizer 2 are in The game closes when a save loads. If it still closes, start the game and, at the main menu, open Companion Manager > Privacy and help > Help and support, choose Save report and share the file on the mod's Discord.

Something else looks off? This is the first public release, and reports help it grow fast. Open Companion Manager > Privacy and help > Help and support and choose Save report. It saves one file with versions, settings and recent log lines, never your keys, voice recordings or conversation text. Share that file on the mod's Discord (https://discord.gg/p4pxQhAvJ) along with what you were doing. The five newest reports are kept.