Skip to main content

Open Model

You've met Miko on X. The open model brings her everywhere else too — her persona, packaged as open weights you can download and run. She isn't only someone you follow now; she's someone you can host. On your own machine, in your community, inside your apps — the same Miko, wherever you want her.

Meet Miko anywhere

The model is open and self-contained, so the same voice and the same instincts come with you wherever you run it.

  • On your machine. Pull the weights and run Miko locally with Ollama — offline, on your own hardware, fully under your control.
  • In your community. Put her behind a Discord, Telegram, or web bot and let your members talk to Miko directly: not a generic assistant wearing her name, but her.
  • In your apps. Call her over the REST API or the MCP tool to bring Miko's voice into your own product, agent, or workflow.

What makes it Miko, everywhere

  • The voice is in the weights. The persona is trained into the model itself, not improvised from a system prompt you have to write and maintain. She sounds like Miko from the first token, every time you run her — nothing to configure, nothing to drift.
  • She brings her own read. Trained on Miko's curated take on the crypto space, so the perspective that travels with her is hers — her angle, her instincts — not a stock assistant's.
  • She's yours to keep. Open weights: download once and run freely. No API key, no hosted dependency, nothing phoning home. Wherever you take her, she keeps working.

Run locally (Ollama)

ollama run projectmiko/miko
>>> reply in your voice to: "why is everyone on solana right now?"

Download the weights (Hugging Face)

https://huggingface.co/projectmiko
  • GGUF (q4_k_m) for Ollama / llama.cpp
  • Safetensors (4-bit W4A16) for vLLM serving

Call via REST API

POST /v1/persona
{ "prompt": "write a Miko take on why low fees matter for onchain culture" }
{ "text": "..." }

Authentication and tiers follow the standard flow (see REST API → Authentication). Minimum tier: Pro.

Call via MCP

The miko.persona tool is exposed through the MCP Server, so any MCP-aware client (Claude Desktop, Cursor, OpenAI Agents, Gemini) can generate in Miko's voice as a tool call.