Everything this app can do, with an example of each. Every item below is a feature that ships today, grouped by the part of the app it belongs to.
Answers appear token-by-token as they're generated, with a stop control while still writing. Ask a question and watch the answer type itself out in real time.
A searchable list of every model across all connected providers, with context size and capability badges. Click the model name in the header, type "llama" to filter, and pick a specific model.
Picks the best available model per message and silently retries another if one fails or is rate-limited. Leave the model on "Auto" and a coding question gets routed to a tool-capable model automatically.
Every chat saves automatically to your browser's local database as you type - no account, no server. Close the tab mid-conversation and come back tomorrow - it's still in the sidebar.
Keep any number of separate chat threads, rename them, or delete individual ones. Keep one thread for "trip planning" and another for "code review."
Several real providers work with zero signup or API key - just add and start chatting. Open Settings, add "Pollinations AI" with one click, and it's usable right away.
When a provider is rate-limited or fails, Auto mode tries the next healthy candidate instead of showing an error. A free provider hits its rate limit mid-session; the next message is silently answered by a different one.
Over 160 open models can be downloaded once and then run entirely on your own device - no network call, no rate limit. Install a small model from the Store, then chat with it with no internet connection.
Chrome's built-in model runs with zero download, useful as an instant fallback. Add the Gemini Nano connection in Settings; it answers instantly with no download wait.
Point the app at your own self-hosted OmniRoute gateway instead of connecting to providers directly. Run OmniRoute on a home server, enter its URL in Settings, and route every chat through it.
Route requests through your own HTTP, HTTPS, SOCKS5 or SOCKS4 proxy, with health-based failover across a pool of them. Useful when a provider is blocked in your region. Add a SOCKS5 proxy in Settings > Proxies, press Test to see which IP you now exit from, and every chat, search and image request leaves from there instead of your own address.
Lets the assistant search the live web and read the full text of top results before answering. Toggle search on and ask about something recent - the assistant cites a page it actually read.
Switch the default search backend to a keyed provider for different result quality. Add a free Serper key in Settings and switch the search backend for more precise results.
A reusable expert instruction pack that can be summoned mid-conversation to change how the assistant behaves. Install a "Code Reviewer" skill and click it in the composer before pasting code.
Install skills from any public GitHub repo publishing the standard marketplace format, with full instructions shown before installing. Browse the bundled marketplace in the Store, preview a skill, and click Install.
The assistant can open URLs, follow links, and read page text through a CORS-friendly reader. Ask "summarize this documentation page" with a URL.
The assistant can run code in any installed language runtime, sandboxed with a timeout, instead of just describing it. Ask for a brute-force computation and get back the real, computed value.
Exact developer utilities the model can call directly - regex, diff, JSON/CSV, hashing, IDs, JWT, color, dates, units. Ask "decode this JWT" and the assistant calls the exact tool.
Register your own REST API endpoints so the assistant can call them as tools. Add a weather API connector, then ask the assistant to look up today's forecast.
A standalone system prompt, model, and toolset that works through a task across multiple steps until done. Build a research agent with browsing tools and give it a question to investigate on its own.
An agent can delegate to other agents, or fan work out to several member agents in parallel and synthesize their results. A research-team agent delegates to specialist sub-agents as needed.
Every agent run shows its full step-by-step trace - not just the final answer - so it can be debugged. Watch each step an agent takes as it runs, laid out in order.
A single catalog of everything installable - language runtimes, utility tools, and local AI models - all installing straight into the browser. Install "Python" from the Store to enable code execution in Python.
Schedule a recurring prompt that runs automatically while the tab is open, recording its result and history. Create a daily task, "Summarize today's top AI news," and check back each morning.
Export chats, skills, memories, and tasks as a single zip and re-import later, merging by ID. Export a backup before switching computers, then import it on the new machine.
Generate images from a text prompt, or edit an existing image with an instruction. Type "a watercolor painting of a lighthouse at dusk" and generate.
Generate short video clips from a text prompt. Prompt "a slow zoom-in on a cup of coffee" and generate a clip.
Transcribe attached audio files to text. Attach a voice memo and ask the assistant to summarize what was said.
The assistant can remember durable, personal facts across conversations and apply them in future chats. Mention once that you prefer metric units; it applies that in later chats too.
Auto-learned facts queue as pending for you to approve, edit, or reject before they become permanent. Confirm or dismiss a suggested memory from the Settings memory tab.
A single cohesive design with an instant, persistent light/dark toggle. Toggle to Dark in Settings for a near-black interface at night.
Downloading InBrowser