Add your API key in Settings, then start a conversation.
Settings
AI Provider
OpenAI API Key
No key entered
Model — enter API key first
Response controls
Each model has its own internal limits. These sliders reflect the widest range any supported model allows — if you select a value that exceeds what your chosen model supports, the model will fall back to its maximum value for that parameter.
Creativity (Temperature)1.00 · Balanced
How chaotic or predictable the replies are
Response Length (Num Predict)10000 · Very Long
How many output tokens AĪRIS is allowed to write per reply
Memory (Context Length)4096
How far back in the conversation it can see
Focus (Top P)1.00 · Balanced
Cuts off unlikely words — lower = tighter, higher = more varied
Advanced settings ▾
Focus (Top P)1.00 · Balanced
Cuts off unlikely words — lower = tighter, higher = more varied
Range (Top K)40
How wide it casts the net when picking words
Floor (Min P)0.05
Minimum certainty a word needs to fit the context before the model can pick it.
Freshness (Repeat Penalty)1.10
How strongly it avoids repeating words it already used
Wandering (Presence Penalty)0.00
Higher values push toward new topics; lower values keep the model on the same topic
Higher values push toward new topics; lower values keep the model on the same topic
Lock (seed)
Pin a number for reproducible results — best-effort, not guaranteed.
Advanced settings ▾
Focus (top_p)1.00 · Very focused
Cuts off unlikely words — lower = tighter, higher = more varied
Range (top_k)50
How wide it casts the net when picking words
Effort (effort)Medium
Reasoning effort for supported models
Extended Thinking (thinking)
Enables extended reasoning. Only supported on Claude Sonnet 3.5 and higher.
Persona (System Prompt)
Service tier — speed & stability
Quick Suggestions — shown on home screen
Instructions
OpenAI
Go to platform.openai.com and sign in or create a free account.
Open the left menu and select API keys.
Press Create new secret key, give it a name, and copy the key. It starts with sk-. Save it somewhere safe — you cannot view it again.
Add a small amount of credit under Billing — even $5 can run for a surprising amount of time depending on how you use your tokens. Start new chats often to drastically reduce cost. An approximate cost guide (based on gpt 5.1) is given in the app though this varies greatly based on the specific model called.
Paste the key above and press Test Key. Then choose your model.
Ollama
Use the command below to start Ollama with browser access allowed, then keep the server URL set to http://localhost:11434.
OLLAMA_ORIGINS=* ollama serve
If you are on Windows, copy and run this exact command instead: set OLLAMA_ORIGINS=* && ollama serve.
Install Ollama from ollama.com and run it locally.
Use the command above to start Ollama with CORS enabled.
Keep the default base server address in the Server URL field above, unless it differs.
Press Test Key to verify the server and load your models.
Claude
Enter your Anthropic API key in the field above.
Press Test Key to verify and load Claude models.
Select the Claude model you want to use from the list.