A workspace that keeps up with your thinking.
Move between models, keep useful context close, and make room for focused work.
One place for your AI conversations
Connect a provider account, choose a model, and start a conversation. Responses appear as they arrive. You can stop a response, copy text, edit your prompt, or regenerate an answer. Each response identifies the provider and model that produced it.
Switch models inside a conversation when another model is better suited to the next question. Earlier messages remain in the conversation and are sent as context to the model you select. When a model cannot accept an existing image or PDF, the workspace asks you to choose a compatible model or remove the file.
Keep related work together
Projects group conversations and carry instructions that apply to their chats. Use them for a client, a research topic, or an ongoing piece of work. Pin conversations you return to often and search titles and message text when you need to find an earlier answer.
Custom assistants save a name, instructions, and an optional preferred model. They are useful for a recurring task such as editing copy, reviewing a proposal, or explaining code. They use the provider accounts you connect.
Compare before you commit
The comparison view sends a prompt to two selected models and displays the responses next to each other. Each response uses its selected provider connection. Two responses mean two provider requests, and each may use paid credits. Compare substance, supported facts, and usefulness for your task.
Bring in supporting material
Upload images when the chosen model supports vision. Native PDF input is available only for supported models and adapters. Plain text, Markdown, CSV and JSON are passed as text. DOCX uploads are converted to text, so layout, charts and embedded images are not preserved. File limits are shown in the workspace.
With tools enabled, a model can propose a text, Markdown, CSV or JSON download. You review the proposed name and content before the file is saved. Generated pictures, audio and video are not part of this release.
Optional context and tools
Memory starts disabled. If you enable it, you can add, review, edit and delete the notes included in future conversations. A model cannot quietly add a memory: proposed notes need your approval.
Tools are also optional. Page reading can retrieve public HTTPS pages. Web search requires the site owner to configure Brave Search. The activity feed shows tool calls and results. Agent runs have a five-step limit and no shell or server administration access.
Your accounts, your choices
Provider credentials are stored encrypted and requests pass through the server. Your browser never receives saved API keys. You can remove connections, download a data export or delete your account. The site owner controls the server and backups, while each selected provider controls its own handling of the requests you send.