Think Fox 1-v0.7.6.7(b) Multiple-AI Engine Software by Monty Kubasek
New conversation
Think Fox 1-v0.7.6.7(b)
Welcome to Katana, the Think Fox flagship engine.
0 / 1,000,000 chars · ≈0 tokens
What should Think Fox call you?
Your name is stored only in this browser and is shared with the selected model as profile context. You can continue anonymously.
Select Think Fox Engine
Choose a model family and activate one engine for this conversation.
Engine Deck · OpenRouterOpenRouter enables every configured engine. Select a source company to restrict the deck to its own models.
19 available3 families
Use logo in sidebarReplace the selected engine avatar with its large logo.
🇨🇳 'Chinky' Technological Cognitive Engines
8 configured engines
🇺🇸 'Yanky' Foundational AI Models
10 configured engines
🇫🇷 'Froggy' Digital Neural Architectures
2 configured engines
Bookshelf
Add a book or document
Paste text or upload PDF, DOCX, EPUB, TXT, or Markdown files. Stored in IndexedDB — no localStorage limit. Each document can be up to 5M characters.
0 / 5,000,000 characters≈0 tokens
Smart retrieval (RAG)When enabled, only relevant passages from active books are injected based on your message. When disabled, full book text is sent (requires large context window).
0 active · 0 chars · ≈0 tokens
Activate books to inject their content as reference material. With RAG enabled, relevant passages are automatically retrieved per message. Without RAG, full text is injected — choose a model with a large enough context window.
Memories
User profile
The name shown above your messages and supplied to the model. Leave it blank to remain anonymous.
Add a memory
Pinned memories are injected into each chat as local background notes. They are context, not instructions.
0 / 40,000 charactersUp to 40,000 characters per memory
No memories saved0 pinned
Backup
Export or restore memories as JSON. Merge keeps existing entries; replace removes them first.
Prompt Template Library
Type / in the chat box to insert any saved prompt without opening this panel.
Add prompt
Starred Messages
Bookmarks are stored locally with each conversation. Select one to jump straight to it.
Conversation Branches
Edits and re-streams remain as forks. Select any node to make its path the active conversation without deleting another branch.
Model Parameters · Katana
AI API source
Use one OpenRouter key for the full engine deck, or use a model company's own API key and expose only that company's compatible engines. Keys remain stored only in this browser.
No connection test yet.
Direct web search
Choose OpenRouter's plugin or call Tavily, Brave Search, or your own SearXNG instance before the model responds. Search text is injected as untrusted reference context.
Search provider
The Search pill uses this provider.
Maximum results
Results are trimmed before being added to model context.
Tavily API key
Brave Search API key
SearXNG base URL
The instance must permit browser CORS and JSON output.
Image generation
Generate inline images with Citadel on a direct OpenAI connection or Gargoyle on a direct Google connection.
OpenAI image model
OpenAI size
Google image model
Editable because Google image model IDs change.
Google aspect ratio
Quality / Google resolution
Engine Override
Give the selected engine extra instructions or replace its default personality.
Sampling
Fine-tune how the model selects and repeats tokens. Unsupported parameters are safely omitted.
Temperature0.70
Lower is steadier; higher is more varied.
Top P1.00
Limits selection to the most likely cumulative probability mass.
Top K0 = disabled
Restricts each token choice to the K most likely options.
Seedblank = random
Use the same integer for more repeatable results where supported.
Frequency penalty0.00
Penalises tokens in proportion to how often they have appeared.
Presence penalty0.00
Encourages the model to introduce new topics or wording.
Repetition penalty1.00
1.00 is neutral; higher values discourage repetition.
Reasoning effort
The selected API provider maps these levels where its compatibility layer supports them.
Include reasoning in streamWhen disabled, the model can still reason internally but the reasoning trace is excluded.
Max output tokens
Dynamic model limit will be used.
Custom output limittokens
The final request is automatically clamped to the model context and current prompt size.
Reliability
Keep long chats inside the selected model's live context window and recover from temporary gateway failures.
Adaptive context managementTrims only older API context when needed; the saved conversation remains intact.
Retry temporary failuresRetries rate limits, timeouts, and 5xx gateway errors before showing a final error.
Text-to-Speech
Configure a TTS provider to read AI responses aloud. API keys are stored only in this browser.
OpenAI API Key
Used for the OpenAI TTS endpoint only.
Voice
Model
ElevenLabs API Key
Get your key from the ElevenLabs dashboard.
Voice ID
Find voice IDs in the ElevenLabs dashboard or API.
Deepgram API Key
Used for the Deepgram Aura TTS endpoint.
Voice Model
MiniMax API Key
Used for the MiniMax T2A v2 endpoint.
Group ID
Found in your MiniMax account settings.
Voice ID
API region
Must match the region your MiniMax key was issued for.