FM26 AI Dialogue mod v1.4 [BepInEx]

Changelog

v1.4

• Added a new openai-compatible Provider option for advanced users to try. Tested with OpenAI and LMStudio (local).
• Altered Google and Openrouter's Auto Model option to use Gemini-3.1-Flash-Lite instead of Gemini-3.1-Flash-Lite-Preview. If you are using either of those as a provider and have a config file from an old version that still lists the models out, you must do the same, or change to use Auto, as the 3.1-Flash-Lite-Preview model will stop working soon.
• Added Deepseek 4 Flash and Pro to Openrouter's Auto Model setting.

v1.3

• Added Personality configuration options to customise your manager: Formality, Ego, Aggression, MediaSavvy and TacticalGranularity, which can all be adjusted from 1-5.
• Or - instead of modifying each personality value independently, you can instead set a single Archetype: TheSpecialOne, TheProfessor, TheHeavyMetal, TheTinkerman, TheOldSchool, TheDiplomat and TheChaosAgent, which override the personality values to specific numbers, with at least one being an extreme value (1 or 5).
• Altered how the Model config values work - you can still set a list of models to use, but the new default is just Auto which uses a list provided by the mod itself (that may change version to version) - this is the safest way to keep up with tested models.
• Added full support for Polish and Spanish languages.

v1.2

• Made the "thinking..." boxes a bit bigger to make them jump around less when the rewritten dialogue is actually put into them.
• Added a new 'Language' configuration value (under [General])to force dialogue to be a particular language regardless of FM in-game user interface setting. Default is 'Auto' which will leave it up to the AI model - sometimes it will get it right, sometimes it will just be in English. Other allowed values are: English, Portuguese, French, Turkish. You'll need to start and quit the game once first after installing 1.2 for the Language item to be put into your config file. Thanks to AsotOne for the idea.
• Very slight tweak to prompts when talking to players and agents (via pop-up window) to try and reduce chances of you referring to them by their full name.

v1.1

• Changed prompts to make team talks and press conference rewrites a LOT more varied. Team talks should be more passionate and motivational, press conferences will be more formal and soundbitey. Maybe I've gone too far but it's certainly more interesting! (see new screenshot)
• First attempt at supporting non-English languages - Portuguese should be fully working, others should work for everything except press conferences. No guarantees the AI models will produce good output, but I've done what I can on my end.
• Fixed team talks for incoming subs

Supported Dialogue Scenes

• One-on-one player or agent conversations (pop-up)
• Player+Agent group meetings (full-screen)
• Team meetings
• Press conferences
• Tunnel interviews
• Team talks

Where possible, extra on-screen context is used to give the AI more to work with when deciding what to say. E.g. what the result was in a post match press conference, or a player's morale, hierarchy position and relationship with you in a one-on-one chat. Obviously the most important thing is the original text, as we don't want it saying things that make no sense or don't help you understand what the real gameplay impact of clicking that option is. A few items of context are carried over to other screens so it gets a little more intelligent the more you play.

Not (yet) supported

• Any other interactions not in the supported list. Examples (will add more to the list as they are discovered):
• Board interactions (not rewritten; original text remains)
• Stuff tucked away in sidebars like during contract negotiations or creating recruitment focuses - will probably never support these, does anyone even read them?!
• Any operating system other than Windows - Linux may work, or may not - completely untested
• In-game languages other than English, French, Portuguese, Turkish, Polish and Spanish - for the rest, most press conferences will not work, other dialogues may or may not.
• Women's football - untested, the rewriting may work but it might assume the players are men unless the original text mentions they're not

Definitely broken

• During a press conference, if you are asked a question where the only option is a dropdown box (e.g. "What position is the player best at") then it breaks rewrites for the next few questions.
• During a press conference, if a journalist re-asks a question ("the public deserve an answer...") then again rewrites get broken for a bit.

STEP 1: Install prerequisites

This is a BepInEx 6 based mod. If you have BassyBoy's stadium mod installed (or any other BepInEx mods) then you already have it and you're good to go straight to step 2.

Otherwise, you need to install it.

1. Get the latest BepInEx v6 nightly from https://builds.bepinex.dev/projects/bepinex_be for your operating system - it should be the "IL2CPP x64" build (for example, at time of writing, if you are on Windows you'd want `BepInEx-Unity.IL2CPP-win-x64-6.0.0-be.755+3fab71a.zip`).
2. Extract it to your FM folder, if you've done it right then then you should see a BepInEx folder appear in there, for example `SteamApps\common\Football Manager 26\BepInEx`
3. Run the game once - if done correctly, you'll see a window pop up alongside the game. Leave it alone.
4. Quit the game after loading is finished.

Where is your FM folder? Common locations are:

Windows - Steam:
C:\Program Files (x86)\Steam\SteamApps\common\Football Manager 26

Windows - Epic:
C:\Program Files (x86)\Epic Games\FootballManager26

Windows - Microsoft store/Game Pass:
C:\Xboxgames\Football Manager 26\Content

Basically it should be the place where the main game executable is, i.e. `fm.exe` on Windows

STEP 2: Install the mod

• Extract the AIDialogue zip file (it should just have one DLL file inside it) into: <FM folder>\BepInEx\plugins\ - You may need to create the `plugins` folder if it does not exist
• .
• Launch FM26 (in order to generate the mod's configuration file)
• Quit the game.

STEP 3: Configure the mod

TL;DR Provider and APIKey

The configuration file can now be found at <FM folder>\BepInEx\config\AIDialogue.cfg. It can be opened with any text editor.

The mod supports three LLM providers. Set Provider in the [General] section to the provider you want to use, then fill in the corresponding APIKey in the section that relates to your chosen provider. You do not need API keys for providers you are not using.

The default provider is groq (link). You can sign up to this and create an API key without providing any payment information, although you can pay if you want higher rate limits. Note as of v1.3 the groq free tier does not get you very much before being rate limited.

Other providers currently supported are openrouter (link), google (link) and openai-compatible (for any OpenAI-compatible HTTP endpoint, including local servers). Which you use is a combination of personal preference, cost and which models are available in each.

Each provider (except openai-compatible) has an Auto Model setting that uses models pre-chosen by me that seemed to work well (low latency, low price, limited hallucinations) in my testing, you can keep these as they are or set your own if you want to experiment.

STEP 4: Go

With the provider and API key (and endpoint, for openai-compatible provider) set, you're good to go, start the game and play.

Optional: Provider info

Groq

Auto Model setting = "llama-3.3-70b-versatile,openai/gpt-oss-120b"

I chose these as they are fairly cheap while still being fast and accurate 'enough'. The free tiers are just enough to evaluate if you want to keep using the mod. They are all quite cheap if you do want to add money to your account. (at time of writing, groq pay tier is unavailable to new users)

The Thinking configuration setting is turned off by default as it uses alot of tokens and you hit the free rate limit much quicker. If you are a paying user, you could turn this on to make the rewritten text more interesting, although it is a bit slower and very slightly more expensive.

Google AI Studio

Auto Model setting = "gemini-3.1-flash-lite,gemini-3-flash-preview"

I chose the most recent Flash and Flash-lite models as they are relatively cheap and very good (maybe the best) at following the instructions - weirdly the pro models tend to hallucinate more as well as being much slower and more expensive - maybe there's such a thing as trying TOO hard! Thinking is turned off by default for the same reason.

As far as I can tell, the free tiers are not high enough for a decent amount of gameplay so you'll need to set up billing - it's pay later rather than pay up front.

HUGE WARNING: Make sure to set a spend cap for your API key!

OpenRouter

Auto Model setting= "openai/gpt-3.5-turbo,openai/gpt-5.4-nano,google/gemini-2.5-flash-lite-preview-09-2025,google/gemini-3.1-flash-lite-preview,google/gemini-3-flash-preview,deepseek/deepseek-v4-flash,deepseek/deepseek-v4-pro"

For this provider, the default set of models is just Gemini , GPT and Deepseek for variety. Also, because Openrouter chooses from many providers, sometimes it picks a really slow one. You can counter that by telling it to prioritise throughput in your routing settings (on the openrouter web site), although that might slightly increase cost.

The Thinking setting is again off by default, I found it slowed things up too much, and it's not really needed for the default models (may be for others). Feel free to experiment.

I haven't tried changing the models to any of the ones they offer for free, but the free rate limits aren't great so paying is probably the best option.

OpenAI-compatible endpoint (advanced users only)

No Auto Model setting. You must list your own!

ApiKey setting is optional for this provider. Leave as `None` if your endpoint does not require auth.

EndpointUrl is a new setting just for this provider. It must be a full chat-completions URL. Example: http://localhost:1234/v1/chat/completions - Tested with LMStudio (Gemma and Qwen models) and OpenAI themselves (gpt-5.4-nano works really well).

Optional: Other config values

Language

Forces dialogue to be a particular language regardless of FM in-game user interface setting. Default is Auto which will leave it up to the AI model - sometimes it will get it right, sometimes it will just be in English or the language of the league you're in. Other allowed values are: English, Portuguese, French, Turkish, Polish, Spanish.

Personality

This lets you shape how "you" sound when the AI rewrites your manager dialogue.

There are two ways to use it:

1. Set the five individual traits yourself (all are 1-5)
2. Pick an Archetype preset (which overrides all five traits)

If Archetype is set to anything other than None, the (hidden) trait values from that archetype are used instead of your manual values.

Individual traits (1-5)

Aggression

• 1 = Stoic and diplomatic, almost never confrontational
• 3 = Firm and competitive, willing to push back
• 5 = Fiery and combative, quick to challenge criticism

In player chats this controls how hard-edged you come across. In media settings it controls how much you bite back.

Formality

• 1 = Very casual, dressing-room style language
• 3 = Professional but natural spoken tone
• 5 = Formal and authoritative, less relaxed phrasing

This mostly affects word choice and sentence style, from "one of the lads" to "proper interview mode".

Ego

1. 1 = Deflects praise to players/fans, takes no credit
2. 3 = Balanced credit-sharing
3. 5 = Strong self-belief, puts success down to your influence

Most noticeable in press conferences and interviews where credit/blame framing matters.

MediaSavvy

• 1 = Brutally honest, minimal filter
• 3 = Cooperative and straightforward, but sensible
• 5 = Polished PR operator, keeps things on-message

Higher values tend to avoid headline-grabbing statements, lower values are more likely to say the blunt thing out loud.

TacticalGranularity

• 1 = Motivation and mentality only (heart, desire, character)
• 3 = Normal tactical language (pressing, midfield control, shape)
• 5 = Full football nerd mode, technical and detail-heavy

Use this to decide how "coachy" your dialogue sounds.

Archetype

• None (default): use your manual trait values as set above
• TheSpecialOne : The ultimate "Mind Games" manager. Self-obsessed and controls the narrative. While a genius, their public persona is more about "heritage" and "respect" than half-spaces and transitions.
• TheProfessor : This manager treats football like a fine art or a complex physics equation. They are calm, slightly detached, and focused on the long-term "project."
• TheHeavyMetal : This manager is all about emotion, "pashun," and high-intensity football. They speak like one of the lads but possess a razor-sharp tactical mind.
• TheTinkerman : The modern tactical obsessive. Not usually aggressive with the media, but they will start dropping terms like "inverted wing-backs" and "high-turnover metrics" in every sentence.
• TheOldSchool : Classic "Brexit Ball" manager. Lots of slang, all about "passion" and "desire," not in depth tactical analysis. Doesn't care about keeping the media happy.
• TheDiplomat : The "Uncle" figure. Never blames anyone, speaks very properly and stays perfectly on message.
• TheChaosAgent : A volatile, emotionally charged manager who oscillates between tactical masterclasses and burning the club down in a post-match press conference.

Archetypes are intentionally opinionated and include at least one extreme value, so each one pushes a clear voice rather than a subtle tweak.

FAQ

What do you use?

In my long term save I use openrouter with model set to Auto, thinking off. Costs a few US cents per hour.

Is this mod free?

Yes. The cost of the AI itself is yours, though.

Will it work on FM24?

No.

Do I need to start a new save?

No.

My dialogues are starting with [rate limited]?

You're using your provider's free quota and have used too much of it - the mod automatically pauses rewriting and lets you know with the [Rate limited]. This is unlikely to happen if you're paying. Depending on which limit you hit, it may start working again in a few minutes, or the next day.

The text isn't changing?

Check that BepInEx is installed correctly into your FM folder and that the mod's `.dll` is in the `plugins` folder. Check `BepInEx\LogOutput.log` for error messages. If it's working you should briefly see "thinking..." in each text box while it does the rewrite.

Text is showing as "AI error" or "thinking..." and there's 401 Unauthorized errors in the log?

Make sure you have replaced `XXX` in the config with your actual API key for the provider you have chosen.

Roadmap

When I have some energy back after this push for v1.0 -

More AI provider support, board interactions, women's football, any other dialogues I've missed and bug fixes.

Removing the mod

If you have no other BepInEx mods, just delete everything that was extracted from the bepinex zip file (4 files, 2 folders) from your FM folder. Otherwise, just delete bepinex/plugins/AIDialogue.dll and bepinex/config/AIDialogue.cfg.

If you just want to temporarily turn it off without deleting anything, you can use `Enabled` in the `[General]` section of the config file. Set it to `false` to fully disable the mod, then restart the game.