Skip to content
OneviumDocs
On this page

Your first task in Onevium

Connect Claude Code or another AI service, choose a project, send a first development task from the composer, follow its progress, approve actions when asked, and check the changes in Review.

Get to know the workspace#

Onevium brings AI conversations, project files, and tools into one window. Your first session follows one path: connect an AI service → choose a project → send a task → follow progress and approve when asked → check the result.

The window has four areas:

AreaWhat it is for
Navigation rail (far left)Home returns to your conversations. Automations, Plugins, Channels, and Images are shown by default; other destinations are under More. The avatar at the bottom opens Account and settings
SidebarNew session, search, Add Project, filters, and conversations grouped by project
CenterThe current conversation and the composer
Right panelOpens with Open panel at the right end of the title bar. The browser, Review, and file previews open as tabs; the terminal docks below the conversation. When a preview needs room, the conversation can shrink to a floating bubble in a corner

This guide uses a fictional development task in a checkout-web project: add promo code validation to checkout.

Install, sign in, and connect an AI service#

  1. Follow the installation guide to install and open Onevium, then sign in to your Onevium account with an emailed code. The Onevium account does not include model credits.
  2. On first launch, with no projects yet, a short setup appears. Click Get started.
  3. If no AI service is available on this computer, you reach Connect an AI service (figure 1):
    • Already signed in to Claude Code in a terminal: choose I already use Claude Code. Onevium checks for a Claude Code sign-in on this computer. If none is found, finish signing in to Claude Code in a terminal, then come back and choose it again.
    • Using an API key or a Coding Plan: choose a provider such as Anthropic, DeepSeek, GLM, Kimi, MiniMax, or Qwen, fill in the form, and click Connect and continue. See provider setup for the credentials each one needs.
    • Not ready yet: click Set up later. Your first send will offer the connection again, and your message is kept.

Figure 1: The connection step of first-run setup. Isolated demonstration of the real 1.3.2 components, simulating a computer with no AI service available. No service was connected and no model was called.

To add or manage providers later, open Settings → AI and connections → AI services.

Choose a project#

A project is the folder for this work; the AI reads and writes files and runs commands inside it. The Choose a project step offers three options:

OptionUse it when
Just startYou have no material yet; Onevium creates a folder in its default location
Open a folderYou already have a code repository or a documents folder, such as checkout-web
Create a new folderYou want to name an empty folder yourself

If this computer has Claude Code CLI conversations, Import from Claude Code also appears so you can continue one of them.

After setup, add more projects with the folder-plus button at the top of the sidebar (Add Project): choose Open an existing folder or type a name to create one. See projects and sessions for conversations and split views.

Send your first task#

Click New session in the sidebar, or hover over a project row and click its pencil button. The home page asks “What are we building today?” The controls around the composer decide where and how the task runs (figure 2):

  1. ① Project: confirm that it shows checkout-web. Click it to search, switch, or Add new project. The same row shows Local and the current Git branch. For a Git repository, the Worktree switch on the right runs the task in a separate directory; leave it off for now and see worktrees later.
  2. ② Permissions: three levels are available. For a first task, choose Default permissions, so actions that need approval, such as running a command, stop and ask you. Auto mode approves safe actions and blocks risky ones. Full access approves every action and asks you to confirm before it turns on. New installations start with Full access; change the global default in Settings → App settings → General → Default Permission Mode. See permissions and planning for the differences.
  3. ③ Model: choose a model from a connected provider, for example Opus 5.5 in the Claude Code group. The same menu sets the reasoning effort.

Figure 2: Check ① project, ② permissions, and ③ model before sending. Isolated demonstration of the real 1.3.2 components; the projects, conversations, and task are fictional and nothing has been sent.

Describe the goal, scope, and how to verify it, for example:

text
Add promo code validation to checkout: call /api/promo/validate when the field
loses focus and show the reason below the field for an invalid code.
Add unit tests first, then change the component.
List the changed files when done, and do not touch payment code.

To point at specific material:

  • Type @ to reference Files (Current Directory or All Files), earlier Sessions, and Capabilities such as the browser or charts.
  • Type / for Commands (such as /init, /review, or /compact) and installed Skills.
  • Click + to add images and files, or switch Mode to Plan so the AI proposes an approach before changing anything.

The tools menu guide covers each menu. Press Enter or click the send button at the bottom right.

Follow progress and approve actions#

After you send, the conversation appears under its project in the sidebar and the AI starts reading and editing files. A collapsed line above the reply summarizes the turn (① in figure 3), such as “Done · 2 edits · 1 read”; expand it to see each tool step.

With Default permissions, an action that needs approval replaces the composer with an approval card (② in figure 3):

  • Read the tool and its details first, such as the command Bash wants to run.
  • Allow Once permits this action only. When a rule suggestion is available, Allow for Session also appears.
  • Deny, or Deny with Feedback with a short reason, and the AI adjusts its approach.

While a conversation waits for you, its sidebar row shows a reminder marker, so you will notice it even while working elsewhere.

Figure 3: ① The turn's progress summary and ② a command waiting for your approval. Isolated demonstration of the real 1.3.2 components with a fictional conversation and command. No command was run and no model was called.

When unsure, deny and explain your concern rather than allowing a command you do not understand.

Check the result#

A reply that changed files shows a file changes card below it, such as “Edited 2 files”, with added and removed line counts:

  1. Click Review on the card to inspect this turn's changes file by file in the right panel. You can also press ⇧⌘G (Ctrl+Shift+G on Windows and Linux) to toggle the review panel.
  2. To discard this turn's changes, click Undo on the card. It previews the files it will restore before you confirm.
  3. To see every file, press ⇧⌘E or choose Browse all files in a new tab of the right panel. To run checks yourself, press ⌘J (Ctrl+J on Windows and Linux) to open the terminal.

In this example, a correct result adds a promo code test file, makes PromoCodeField call the validation endpoint when the field loses focus, passes the tests, and leaves payment code unchanged. If the diff includes files you did not expect, say so in the conversation and ask the AI to explain or revert them. See files, terminal, and Review for reviewing, comparing branches, and committing.

Choose your next tool#

What you want to doGuideExample request
Read documents, edit code, or run checksFiles, terminal, and Review“Update this explanation and list the actual changes.”
Try changes in a separate directoryWorktrees“Confirm the working directory and branch before you start.”
Read a website or check its controlsBrowser“Open the local preview in the built-in browser and check the promo code field.”
Reference files, conversations, or capabilitiesTools menuType @ in the composer
Repeat work using a defined methodSkills“Use this review Skill to check the README.”
Split a complex requirement across conversationsTeam sessions“Split this into development and testing. First list responsibilities and acceptance criteria.”
Run on a schedule or start from team chatScheduled runs and channels“Check this report every day.”

Troubleshoot your first task#

SymptomNext step
“Connect an AI service to send” appears after you sendChoose I already use Claude Code or a provider, fill it in, and click Connect and send. The message stays in the composer and is sent once you connect
You chose I already use Claude Code, but no sign-in was foundFinish signing in to Claude Code in a terminal, then try again in Onevium; or use an API key, see provider verification
The project picker shakes and opens when you sendNo project is selected. Choose one, or click Add new project
No reply, authentication failure, or an unavailable modelCheck the connection in Settings → AI and connections → AI services and test a one-sentence reply; see provider verification
No approval card appeared and the AI edited files or ran a commandCheck the permission level below the composer. Full access approves every action; with Default permissions, edits to project files usually do not ask each time
Files come from another projectCheck which project the conversation is under in the sidebar, and start a conversation from the correct project
Review is missingReview works with Git repositories; use Browse all files in a plain folder

What to do next#

Finish connecting the providers you use (provider setup), manage real work with projects and sessions, and give conversations clear titles and labels with conversation organization. To keep an eye on conversations waiting for approval and today's usage, open the menu-bar hub.