Models & Providers

Every Chat Node runs against a model you choose. The desktop app gives you three ways to run AI, and you can mix them freely across a canvas.

---

Three ways to run AI

1. Rabbitholes managed models (built-in)

No setup required — these use your signed-in session and consume credits from your account. In the model dropdown they appear as tiers:

• Basic — Claude Haiku 3.5 or equivalent
• Pro — Claude Sonnet 4.5 or equivalent
• Advanced — Opus 4.7 / Gpt 5.5 equivalent

2. Bring your own key (BYOK) cloud providers

Add your own API key for OpenAI, Anthropic, Google, Perplexity, xAI, OpenRouter, Straico, Qolaba, DeepSeek, Groq, or Fal AI (Beta). Requests go directly through your key, so no credits are used. Set these up in Settings → Providers.

BYOK providers require a Pro plan or higher.

3. Local models via Ollama

Run models entirely on your machine — free, private, and works offline. See Settings → Ollama.

Two more options extend the list:

• Agent providers — use the Cursor or Claude Code CLIs installed on your machine as chat backends. See Settings → Agent Providers.
• Custom providers — connect any OpenAI-compatible endpoint. See Settings → Custom Providers (Pro plan or higher).

---

The model dropdown

Click the model name at the bottom left of any chat input to open the model dropdown.

1. Models are grouped by source: Rabbitholes tiers first, then agent providers, then your configured cloud providers

2. Type in the search field to filter models by name

3. Click a model to select it for this node

On the free plan, models that require a paid plan show a lock icon and a Paid tag.

Use the Manage AI Models link at the bottom of the dropdown to jump straight to provider settings.

---

Per-model settings

You can tune each model individually:

1. Open Settings and select the provider

2. Click a model in the provider's model list

3. Adjust the available fields:

   • Alias — a display name of your choosing
   • System Prompt — instructions prepended to every conversation with this model
   • Advanced — Temperature (0–1), Top K (1–20), Top P (0–1), Presence Penalty, Frequency Penalty, Max Tokens, Max Steps

4. Click Save changes, or Restore defaults to reset

---

Reasoning effort

For models that support extended thinking, you can control how long the model reasons before answering:

1. Open the + tools dropdown in a chat input (or type / for the slash menu)

2. Select a Reasoning Effort level: Low, Medium, High, or Max

The default is Medium. Higher levels give the model more thinking time for complex problems. See Slash Commands for the / shortcuts.

---

Default models

Set app-wide defaults in Settings → Default Models:

• Default Chat Model — used by new chat nodes
• Title Generation Model — names your conversations; pick a lightweight low cost model
• Text-to-Image and Image-to-Image — used by Image Nodes

You can also override the default chat model per canvas in the canvas settings — nodes inheriting the app-wide setting show a Global default badge.

---

Credits

Rabbitholes managed models consume dollar-denominated credits from your account balance. BYOK providers and Ollama never touch your credits. Check your balance and plan in Settings → Account.

---

Use cases

• Zero-setup start: sign in and use the Rabbitholes Pro tier immediately, without hunting for API keys.
• Cost control on heavy workloads: add your own Anthropic or OpenAI key so long research sessions bill your provider account instead of credits.
• Private offline work: run a local Ollama model for sensitive documents that must never leave your machine.
• Fast titles, smart chats: set a cheap model as the Title Generation Model while keeping an Advanced-tier model for the actual conversation.
• Per-project tuning: give a coding canvas its own default model and a lower temperature via per-model settings, while your writing canvas inherits the global default.