A complete guide to every feature in Inversify AI. Whether you're chatting with the latest models, generating images, searching the web, or managing your account — it's all here.
Sign up with your email or Google account. Free accounts get access to lightweight models (GPT-4.1-mini, GPT-4.1-nano, GPT-4o-mini) with a daily Spark allowance. Paid plans unlock all 70+ models with higher limits.
After signing in, you land on the chat page. Type a message, pick a model, and hitEnterto start a conversation. That's it.
Every message thread is a conversation. They appear in the left sidebar, sorted by most recent activity.
Just start typing in the input bar. Your first message creates a new conversation automatically. An AI-generated title appears after the first exchange.
Pin important conversations to keep them at the top. Rename them by clicking the title. Folders let you group conversations by topic — drag and drop or use the right-click menu. Archiveconversations you're done with but don't want to delete.
Hover over any message to see actions: Copy the text, Edit your message and regenerate the response, Regeneratethe AI's response with the same or a different model, Delete a message, or Branch the conversation from that point into a new thread.
Use the /export slash command to download the current conversation as a Markdown file. You can also export from the conversation info panel (click the info icon in the top-right).
Inversify connects to 5 AI providers — OpenAI, Anthropic, Google, xAI, and Fireworks — giving you access to 70+ models through a single interface. Click the model name in the input bar to open the model selector.
Each model has capability badges:
Models are categorized as free, low, medium, or high cost. The model selector shows estimated Spark cost per message. Cheaper models are faster but less capable; premium models are slower but more intelligent.
You can change models at any time during a conversation. Each message is generated by whatever model is currently selected. The model badge next to each AI response shows which model produced it.
Web search gives the AI access to current information from the internet. Click the search dropdown next to the model selector, or use /search or /research commands.
When web search is active, the AI cites sources inline as [1], [2], etc. A Sourcessection appears at the end of the response with clickable links. You'll see a blue globe indicator while the search is running.
Search works best when your question is specific. “What happened with OpenAI today?” gives better results than “Tell me about AI.” Extended mode shines for multi-faceted questions like “Compare the pricing and features of the top 3 cloud providers.”
Generate images with DALL-E 3 or DALL-E 2. Select a DALL-E model from the model picker, or click the Image button, or type /image.
Type a description of the image you want and send it. You'll see a purple “Generating image...” indicator with an elapsed timer. The image typically takes 5-15 seconds to appear.
Expand the advanced options (slider icon) to toggle between Standard and HD quality. HD produces more detailed images but costs twice as many Sparks and requires a Pro or Max plan (BYOK users are exempt).
Be specific about style, composition, and mood. “A watercolor painting of a beagle puppy sitting in autumn leaves, warm golden light” works much better than “beagle.” DALL-E 3 may refine your prompt — the revised prompt appears below the image.
Hover over a generated image to see the Open full size and Download buttons. You can generate multiple images in the same conversation.
Attach files to your messages by clicking the + button, dragging and dropping files onto the chat, or pasting images from your clipboard.
When you attach an image to a vision-capable model (marked with vision), the AI can see and analyze the image. Non-vision models receive a text note that an image was attached.
Upload files in Settings → Files to make them available as context in every conversation. The AI can reference these files when answering your questions.
You can upload up to 20 files. Each plan has a character limit for how much context is injected (Free: 20K chars, Premium: unlimited up to ~80K chars). Toggle individual files on/off to control what's active. A context usage meter shows how much of your limit is used.
Reasoning-capable models (marked with reasoning) can show their step-by-step thought process. Click the Thinking button to enable it.
When enabled, the AI produces a collapsible “Thinking” block showing its internal reasoning before the final response. This is useful for complex math, logic, and coding problems where you want to verify the AI's reasoning chain. Thinking tokens count toward usage but produce more accurate answers.
Send the same message to multiple models simultaneously and see their responses side by side. Click the Compare button (desktop only) to enable it.
Add up to 2 additional models alongside your primary model. Type a message and all selected models respond in parallel. Useful for evaluating which model works best for your use case.
Click the microphone icon to dictate your message. Speech is transcribed using OpenAI Whisper and automatically sent. Your browser will ask for microphone permission the first time. Voice input uses Transcription Sparks (separate from chat Sparks).
Type / in the message input to see available commands:
| Command | Description |
|---|---|
| /clear | Start a fresh conversation (same model) |
| /image [prompt] | Switch to DALL-E 3 and generate an image |
| /summarize | Ask the AI to summarize the conversation |
| /export | Download the conversation as Markdown |
| /search | Enable Standard web search |
| /research | Enable Extended (deep) web search |
| /translate [lang] | Translate text to another language |
| /help | Show all available commands |
Use arrow keys to navigate the command palette, Enter or Tab to select.
| Shortcut | Action |
|---|---|
| Enter | Send message |
| Shift + Enter | New line (without sending) |
| Cmd/Ctrl + Shift + C | Copy last AI response |
| Cmd/Ctrl + Enter | Send message (alternative) |
Set persistent instructions in Settings → Preferences. These apply to every conversation — great for things like “Always respond in Spanish” or “I'm a senior developer, keep explanations technical.”
Set an AI persona to change the AI's identity entirely. The persona replaces the default system prompt. Use this for role-playing, brand voices, or specialized assistants.
Click the sliders icon in the input bar → “System prompt” to set instructions for just the current conversation. Useful for one-off tasks that need specific behavior.
Assistants are pre-configured combinations of a system prompt and a model. Pick one from the assistant picker in the advanced options. Community-created assistants are available in theCommunity section.
Access pre-written prompts from the “Prompts” button in the advanced options. These fill in your message input with templates for common tasks.
Projects group conversations and knowledge files together. Create a project in the Projects section, add a project-level system prompt, and upload project-specific files. All conversations within a project share the same context.
Access settings from the sidebar or navigate to /settings.
Sparksare Inversify's usage currency. Every AI interaction costs Sparks based on the model used and message length.
Add your own API keys for OpenAI, Anthropic, Google, xAI, or Fireworks in Settings → API Keys. When a BYOK key is set for a provider, all requests to that provider use your key instead of Inversify's — meaning zero Spark cost for those models. You pay the provider directly at their rates.
Keys are encrypted at rest and never logged. BYOK users are also exempt from HD image quality restrictions.
Inversify can remember things about you across conversations. Say “Remember that I prefer Python” or “Don't forget my cat is named Mochi” and the AI will save it. These memories are injected into every future conversation.
Manage saved memories in Settings → Memory. You can delete individual memories or disable the feature entirely. Up to 100 memories are stored, with oldest automatically removed when the limit is reached.