> ## Documentation Index
> Fetch the complete documentation index at: https://help.harmonica.chat/llms.txt
> Use this file to discover all available pages before exploring further.

# Bring Your Own Model (BYOM)

> Use your own AI API keys to power Harmonica with any supported model

## Overview

Bring Your Own Model (BYOM) lets you connect your own AI provider API keys to Harmonica. When BYOM is active, your sessions use your own keys instead of Harmonica's — removing usage limits and giving you full control over which models power your facilitation.

## Why use BYOM?

* **No credit limits** — BYOM sessions don't consume your Harmonica balance
* **Model choice** — pick the exact model for your use case (faster, cheaper, higher quality, or open-source)
* **Data control** — requests go directly to your chosen provider under your own terms of service
* **Self-hosting** — combine BYOM with a self-hosted Harmonica instance for fully self-contained deployment

## Supported providers

Harmonica supports API keys from these providers:

| Provider              | Example models                             | Notes                             |
| --------------------- | ------------------------------------------ | --------------------------------- |
| **OpenAI**            | GPT-4o, GPT-4o mini, GPT-5                 | Most widely used                  |
| **Anthropic**         | Claude Sonnet 4.6, Claude Haiku            | High quality facilitation         |
| **Google**            | Gemini 2.5 Flash, Gemini 3.1 Pro           | Good balance of speed and quality |
| **Ollama**            | Llama, Mistral, Qwen                       | Local models, fully offline       |
| **OpenAI-compatible** | Any provider with an OpenAI-compatible API | Custom base URL supported         |

## LTD plan requirement

If you are on the Lifetime Deal (LTD) plan, BYOM is required to create sessions. LTD accounts do not use Harmonica's monthly AI credit balance — all LLM calls must go through your own keys. If you try to create a session without any BYOM keys configured, the session creation flow will block with a prompt to set up your keys first. A banner on the dashboard also reminds you when no keys are detected.

## Setting up BYOM

<Steps>
  <Step title="Go to Settings">
    Navigate to **Settings** → **Integrations** in your Harmonica dashboard. The **AI Models** section at the top of that tab is where you manage your BYOM keys and model assignments.
  </Step>

  <Step title="Add an API key">
    Click **Add API Key** and enter:

    * **Provider** — select from the supported list, or choose "OpenAI-compatible" for custom endpoints
    * **API Key** — your key from the provider's dashboard
    * **Base URL** — auto-filled for standard providers, customizable for self-hosted or compatible APIs
    * **Name** — a label to help you identify this key
  </Step>

  <Step title="Select models for each tier">
    Harmonica uses three model tiers for different tasks:

    * **Small** — lightweight tasks (e.g., classification, short responses)
    * **Medium** — main facilitation conversations
    * **Large** — summaries, synthesis, and complex analysis

    Choose which of your available models to use for each tier.
  </Step>

  <Step title="Create sessions as usual">
    Once configured, all new sessions automatically use your models. No other changes needed — the facilitation experience is the same.
  </Step>
</Steps>

## Platform models vs BYOM

Even without BYOM, Harmonica provides platform models:

| Model             | Provider            | Tier                   | Available on Free plan |
| ----------------- | ------------------- | ---------------------- | ---------------------- |
| GPT-4o mini       | OpenAI              | Small                  | Yes                    |
| Gemini 2.5 Flash  | Google              | Small / Medium / Large | Yes                    |
| Apertus 70B       | PublicAI (Swiss AI) | Small / Medium         | Yes                    |
| Claude Sonnet 4.6 | Anthropic           | Medium / Large         | Pro only               |
| GPT-5             | OpenAI              | Medium / Large         | Pro only               |
| Gemini 3.1 Pro    | Google              | Medium / Large         | Pro only               |

BYOM gives you access to any model from any supported provider, regardless of your Harmonica plan.

## How billing works with BYOM

* **BYOM users pay their AI provider directly** — Harmonica doesn't charge for LLM usage
* **Session creation doesn't reserve balance** — no credit checks when creating or running sessions
* **Summary generation skips settlement** — no per-participant costs deducted
* You still need a Harmonica account (free plan is sufficient with BYOM)

<Tip>
  BYOM is especially useful for organizations that already have enterprise agreements with AI providers, or for self-hosters who want to run Harmonica with local models via Ollama.
</Tip>