Use a local GGUF
Choose a model already on your machine or download one from Hugging Face. Mirid estimates memory use before you load it.
Reason · Trust · Fidelity
AI, without the ceremony.
Mirid makes capable AI feel legible. Run models on your own machine, use the providers you already trust, and give characters a voice—all without becoming your own systems administrator.
Windows 10 or 11 No GPU required No Mirid account
Local or hostedYour hardware, your APIs, or both
Character-firstCards, scenes, voices and continuity
OpenAI-compatibleUse Mirid as a backend for other apps
Open sourceLicensed under AGPL-3.0
One room, many models
Start where you are.
A gaming PC, an ordinary laptop, or a handful of API keys are all valid beginnings. Mirid shows you what is available and keeps the machinery out of the way until you want it.
Bring an API key
Connect NanoGPT, OpenRouter, OpenAI-compatible endpoints and supported frontier providers. Their models appear in one navigable library.
Use both
Keep intimate conversations local, reach for hosted models when they suit the task, and change course without rebuilding your workspace.
Character has consequences
More than a name and a prompt.
Build characters through conversation, import familiar card formats, stage group scenes and carry the details that make a story feel continuous.
- Agentic character creation that asks useful questions
- Character cards, personas and multi-role conversations
- Per-character voices, portraits and story state
- Conversation, call and distraction-free focus modes
A clean first launch
Install once. Then begin.
The installer stays small. On first launch, Mirid retrieves its versioned runtime, verifies it and sets it up for you. The download resumes if your connection drops.
Step one
Install Mirid
Run the Windows installer. It places the desktop app and the shortcuts it needs.
Step two
Let the engine arrive
Mirid downloads roughly 3.1 GB once, checks its exact size and SHA-256, then extracts it locally.
Step three
Choose your way in
Add an API key, load a GGUF model, or do both. An NVIDIA GPU makes local work faster; it is not required to open Mirid.
Plan for the first launch. Keep about 14 GB free for the verified download and extracted runtime. Local model files need additional space.
Read the first-run guide →
Voice, vision, reach
Let the conversation leave the textbox.
Speak to characters, hear their replies, bring images into the conversation and generate new ones. Use Mirid directly or let another application call its OpenAI-compatible server.
Local GGUFNanoGPTOpenRouterOpenAI-compatibleKokoro TTSSpeech recognitionImage generationSillyTavern
Before you install
Straight answers.
Mirid is ambitious software, but it should never be mysterious about what it needs or where your data goes.
Do I need an NVIDIA GPU?
No. Mirid opens and works on CPU-only Windows machines, and API-only use needs no dedicated GPU. Local text and media generation may be much slower without compatible acceleration.
Does Mirid include an AI model?
The core runtime includes the application engine and a local text-to-speech voice, not a general-purpose chat model. Choose a GGUF model in Mirid or connect a hosted provider.
Will my conversations leave my computer?
Local inference stays on your machine. When you choose a hosted provider, the content needed for that request is sent to that provider under its own terms and privacy policy.
Why is there a 3.1 GB first-launch download?
It contains Mirid’s Python runtime and native inference dependencies. Keeping it outside the installer makes the desktop app smaller and lets the engine receive independent, verified updates.
Can I use Mirid with SillyTavern?
Yes. Mirid can stream text through an OpenAI-compatible endpoint, and its bridge extends that connection to selected voice, speech and image capabilities.
Mirid 1.0 for Windows
Your models. Your characters. One place to meet.
Download the installer, give the first launch a little time, and shape Mirid around the way you actually use AI.