OpenRouter or direct providers
Use one OpenRouter key across the full engine deck, or connect supported source-company APIs. Engines unavailable under the selected provider are clearly locked rather than failing silently.
Think Fox 1-v0.7.7.6 · AI by Monty Kubasek
Think Fox 1
Twenty-one AI engines. One command interface. Route work through OpenRouter or supported source-company APIs, organise conversations into Projects with searchable context retrieval, connect GitHub repositories directly, switch models without rebuilding your workflow, and listen to responses through your choice of professional text-to-speech provider.
Current project
A working AI command centre, not a model picker with decoration.
Think Fox combines multiple providers with persistent conversations, Projects, GitHub repo access, project context retrieval, reusable memory, direct search, document handling, image generation, multi-provider text-to-speech, and branching workflows inside one browser-based interface.
Projects and Workplaces
Organise conversations into Projects inside Workplaces. Assign, move, filter, and archive. Each Project holds its own memories, context index, retrieval mode, and optional GitHub connection.
GitHub repo access
Connect a Project to a GitHub repository using a fine-grained PAT. Browse the tree, sync files into project context, generate patches, and commit updates with SHA-guarded diff confirmation. Tokens never leave your browser.
Project context retrieval
Index conversations into searchable chunks. On each message, Think Fox scores and retrieves relevant context from other Project conversations, memories, and synced GitHub files — within a strict token budget.
Text and PDF context
Upload text, code, data files, or PDFs. Think Fox extracts readable content locally in the browser and attaches it to the relevant message without requiring a document server.
Direct web search
Use OpenRouter search or query Tavily, Brave Search, or a configured SearXNG instance. Results are added as labelled reference context before the engine responds.
Image generation
Switch into image mode on supported OpenAI and Google routes. Generated images are rendered inline and stored with the conversation in the local Think Fox library.
Multi-provider text-to-speech
Play assistant responses aloud using Deepgram, OpenAI, ElevenLabs, or MiniMax. Choose the provider, voice, model, speed, and output settings that suit the task.
Memory, prompts and bookshelf
Keep a user profile, pinned memories, Project memories, reusable prompt templates, and reference documents. Context can be searched, categorised, backed up, and injected only when useful.
Persistent conversation forks
Edit a message without destroying what followed. Think Fox preserves the original path, creates a new branch, and lets you move through the conversation tree.
Diagnostics and repair
Built-in diagnostics panel shows system state, storage estimates, stale items, and orphan records. One-click repair tools fix broken references and export emergency backups.
Workplace export and import
Export entire Workplaces — conversations, Projects, memories, context index, canvas, attachments, and settings — as a single JSON file. Import into any browser. GitHub tokens are never exported.
Engine deck
Twenty-one engines organised into five divisions.
Each engine keeps its own callsign, colour, model mapping, capabilities, default parameters, and concise system prompt. Switch engines within the same conversation whenever the work changes.
Command Core
Fast response, general reasoning, and everyday command engines
Configured
Shuriken
MoonshotAI Kimi K2.6Fast general reasoning, coding, drafting, image input, and everyday tasks.
Default
Katana
MoonshotAI Kimi K3Large-context general work with preserved reasoning, vision, and tools.
Configured
Tempest
Z.ai GLM 5.2Deep reasoning, coding, and tool-assisted work for difficult multi-step problems.
Configured
Cinder
MiniMax M3Large-context reasoning, coding, tools, and multimodal input.
Vector Wing
Technical precision, coding, and structured analysis engines
Configured
Scythe
Qwen 3.7 MaxExecutive analysis, structured writing, coding, and agent workflows.
Configured
Hellblade
Qwen 3.8 MaxSharp technical analysis, coding, agents, and multimodal input.
Configured
Typhoon
DeepSeek V4 Pro 0813Rigorous technical reasoning, precise analysis, coding, and agents.
Configured
Trident
Poolside Laguna S 2.1Specialist code generation, technical problem solving, and long-context engineering work.
Configured
Glaive
Tencent HY3Efficient reasoning, coding, multilingual work, and tool calling.
Strike Squadron
Frontier proprietary reasoning, multimodal, and heavy-lift engines
Configured
Orbit
Anthropic Claude Opus 4.8Nuanced reasoning, long-form writing, coding, images, and files.
Configured
Rocketship
Anthropic Claude Opus 5Advanced reasoning, knowledge work, coding, images, and files.
Configured
Ancestral
Anthropic Claude Fable 5Natural writing, precise analysis, coding, images, and files.
Configured
Citadel
OpenAI GPT 5.6 Terra ProGeneral reasoning, coding, tools, vision, files, and supported image generation.
Configured
Stronghold
OpenAI GPT 5.6 Sol ProHigh-end OpenAI reasoning, coding, multimodal work, files, and tool use.
Oracle Wing
Deep knowledge, multimodal, and creative reasoning engines
Configured
Mudcake
Meta Muse Spark 1.2Deep multimodal reasoning, huge context, mixed media, and dense analysis.
Configured
Gargoyle
Google Gemini 3.1 Pro PreviewLarge-context multimodal reasoning, image understanding, files, audio, video, and supported image generation.
Configured
Eight Ball
xAI Grok 4.6Creative and unconventional reasoning with vision, files, and tool support.
Configured
Neon
NVIDIA Nemotron 3 Ultra 550B A55BHeavy reasoning, planning, and agent orchestration.
Hammer Division
Specialist, multilingual, and open-model engines
Configured
Monolith
Mistral Large 2512Multilingual reasoning, vision, files, coding, and tools.
OpenRouter only
Dagger
CognitiveComputations Dolphin Mistral 24B Venice EditionDirect, creative open-model work with a compact general-purpose footprint.
OpenRouter only
Harlequin
NousResearch Hermes 3 Llama 3.1 405BHigh-capacity open-model reasoning, creative work, and flexible voice.
Provider-aware routing
Use the route you actually pay for.
OpenRouter mode exposes the complete twenty-one-engine deck through one account. Direct-provider mode uses the selected source-company API and automatically locks incompatible engines. Dagger and Harlequin remain OpenRouter-only.
Each provider adapter handles its own endpoint, authentication, request format, streaming events, usage data, and supported capabilities. Think Fox keeps the interface consistent while the routing layer deals with the differences underneath.
OpenRouterFull twenty-one-engine deck, unified model catalogue, pricing and routing.
Direct APIsOpenAI, Anthropic, Google, xAI, Moonshot, Alibaba, DeepSeek, Z.AI, MiniMax, Mistral, NVIDIA and Tencent.
GitHubFine-grained PAT connection, repo tree browsing, file sync, SHA-guarded writes. Token stays local.
SearchOpenRouter plugin, Tavily, Brave Search or configurable SearXNG.
Text-to-speechDeepgram, OpenAI, ElevenLabs and MiniMax voice routes with provider, voice, model and playback controls.
Local dataConversations, Projects, context index, memories, documents, GitHub files and generated images remain in browser storage.
ShurikenKatanaTempestCinderScytheHellbladeTyphoonTridentGlaiveOrbitRocketshipAncestralCitadelStrongholdMudcakeGargoyleEight BallNeonMonolithDaggerHarlequinDifferent models are good at different work. Think Fox makes switching deliberate, visible, and useful.One interface · twenty-one engines · five divisions · your API routes
Think Fox 1 is ready to launch.
Bring an OpenRouter key for the complete twenty-one-engine deck, or connect supported direct-provider and text-to-speech accounts. Organise work into Projects, connect GitHub repositories, retrieve context from past conversations, generate images, search the web, work with documents, and listen to responses aloud—all from the same interface. Keys, tokens, and project data are stored locally in your browser, so use Think Fox on a trusted device.
Launch Think Fox →