New to Onevium? Start with these 10 everyday tips

Switching between requirements and a draft makes it easy to lose your place. Put both conversations side by side, keep the requirements visible, and compare as you edit. Two approaches are easier to judge when you can read them together.

Once your model is connected, give split view a try. The other tips cover models, child conversations, titles, colors, and memory. Read whichever one you need next.

Split view saves you from switching back and forth. Keep the requirements on the left and edit a draft on the right, then compare directly. If you have two possible approaches, put them next to each other and read both without losing your place.

Drag another conversation from the sidebar into the current chat area. Wait for a left or right split target, then release. Drag the divider to give the longer conversation more room.

You can also keep several ongoing tasks in view to see which is still working and which has a result. When using child conversations, open the lead and members alongside each other. On a wide screen, drag in another conversation when you need it.

To compare writing from different models, select a model in each conversation and give both the same text and instructions. Read the results side by side, then continue in the column with the version you prefer.

Before sending a message, click the intended column and check its title and composer. Closing a column only closes that view; the conversation remains in the sidebar. Close the extra columns when you are done comparing.

Conversations pointing at the same directory still share its files. For code edits, give them separate file responsibilities or use worktrees. See projects and conversations for the split-view steps.

If you use Claude Code, sign in through its connection status panel. For a provider API key, open Settings → Providers → Add AI service and choose the appropriate preset. OpenAI, Grok, Gemini, and DeepSeek use their respective accounts; a team gateway goes under Custom service.

Start with one connection. Save it, select it in the chat composer's model menu, and ask: ‘Confirm receipt in one sentence. Do not read files or run commands.’ An actual reply confirms the connection works. Check configuration only inspects local settings; it does not contact the provider.

To change models later, use the composer. Each conversation keeps its own selection. You do not need to delete a connection to use another model. Claude subscription sign-in and API keys are separate connection methods. See connecting providers for the full steps.

A website edit and a quote review are easier to find later when they live in separate chats. You can return to the relevant task without searching through a long conversation about several unrelated things.

Create the conversation under the right project, then check the project and working directory before asking it to read files. This is especially useful when two projects have similar names.

One website project might have ‘Rewrite homepage intro,’ ‘Check mobile navigation,’ and ‘Update FAQ.’ The project keeps related work together; each chat gives one task somewhere to continue. See projects and conversations.

Some jobs have useful divisions: one member prepares a quickstart, another collects common questions, and a lead checks the two documents together.

For ongoing collaboration like this, select Team in the composer's @ menu. First confirm that your environment has the matching team Skill and that Settings → Tools has the Sessions tools enabled. Describe what each member can read and write, have the lead propose the assignments, and start the real member conversations under the confirmed plan.

Members appear beneath their parent in the sidebar, and you can open their individual chats. A useful assignment is ‘One member writes only quickstart.md; the other writes only faq.md; the lead reviews both.’ Conversations using the same directory still share its files. See Team collaboration before starting.

When you have a version you want to keep, branch instead of repeatedly rewriting the same conversation. Find a completed assistant reply, choose Branch to new chat beneath it, and confirm Create new conversation.

The new chat keeps the conversation up to that reply. You can ask, ‘Keep the earlier requirements, but make this version shorter,’ then compare it with the original.

This works well for writing and exploring ideas. For independent code changes, use a worktree: branching a chat still uses the same workspace and does not copy the project files.

‘New conversation’ and ‘Take a look’ tell you very little a few days later. Right-click a conversation, choose Rename…, and try something specific, such as ‘Investigate duplicate payment callbacks.’

You can also choose AI generate, check the suggestion, and save it. During an active task, rename manually or wait until the task finishes to request an AI suggestion. Automatic naming respects titles you saved manually.

For more consistent titles, open Settings → Conversation organization → Title rules. Select a naming model and give it a short instruction: ‘Describe the task, keep the important object, and omit greetings.’ This model is separate from the chat model. Enabling automatic naming does not immediately rename all historical conversations.

Right-click a conversation → Mark to choose a color, emoji, or existing tag, then save the mark. It helps distinguish work waiting for review from work you have simply set aside.

Create a reusable ‘Needs review’ tag under Settings → Conversation organization → Tags. Apply it to relevant chats and filter by it later. Colors help you spot a task; tags help you collect related ones.

A green mark does not verify that the work is complete. Open the result and check it. If a conversation seems missing, clear search and filters first. See conversation organization.

Open Settings → Appearance → Theme and fonts. Choose light, dark, or system mode, then a color palette. Interface, message, and code fonts can be adjusted separately.

Check a real conversation with long text and code instead of relying only on the settings preview. If it feels cramped, try the spacious density in Appearance → Conversation display. Use compact density when you want more messages on screen.

Theme and font changes apply across conversations. Use conversation marks to distinguish individual tasks. See settings and data for the other options.

If you keep asking for conclusions first, less small talk, or examples in a particular language, put a few clear instructions in Settings → Personalization → User rules. Stable preferences are easier to maintain than a long collection of temporary requests.

For established project facts, use Memory at the bottom of Personalization. Check the project scope before saving. Keep project-specific agreements out of global preferences, and correct old entries when an agreement changes. If a new project is absent from the filter, work in it with extraction enabled until it has project memories; do not save its facts globally as a shortcut.

The @Memory request helps with rule and memory files; it is separate from this memory list. Try a relevant request in a new conversation to check how the information is used. Saving an entry does not guarantee recall on every turn. See memory and recaps.

When you can see several changes you want to describe, speaking may be easier than typing. Enable voice input in Settings → Voice, select your microphone, download and activate a recognition model, then use the composer's microphone.

Try: ‘Shorten the opening to two paragraphs. Keep the dates and numbers. Leave the rest alone.’ Check numbers, filenames, and restrictions such as ‘leave alone’ before sending. Recognition behavior depends on the selected model.

Speech recognition runs locally. Optional AI polish sends the transcript text to your selected model; turn it off if you want the raw transcription. If sending immediately after dictation is enabled, the transcript goes straight to the conversation. Turn that option off when you want to read it first. See voice dictation.

If your model is connected, drag two related conversations into a split view: keep the requirements in one and the result in the other. Give them recognizable titles and save one preference you keep repeating. Those are three things you can use today.

The task names and filenames above are illustrative examples. Replace them with your own work. If you are still setting up your first connection, continue with getting started.