Start with Tvijo in ten minutes

You can be working inside Tvijo before you read the rest of this wiki: open the workspace, sign in, ask one question, then look at the receipt before you ask a second one. That last step matters — Tvijo is built so that every task has a visible price and a visible limit.

🧭 What you get in this first session

🧭 Step 1 — open the workspace

Tvijo AI is the workspace brand. Centavra is the simple browser entry and DSHREM is the technical and editor entry; both lead to the same account and the same task rules. The manual at dsh.tvijo.com/learn is the canonical starting page.

🧭 Step 2 — sign in without pasting a secret

The sign-in flow does not ask you to copy an API key into a form:

1. The page puts an eight-character code on your clipboard.
2. Open browser opens a prefilled approval link.
3. You sign in with Google or a password, then you return to the approval form.
4. Approval is still explicit — nothing is granted until you confirm.

Treat the code like a one-time password: anyone with it and the approval page can finish the flow. Do not paste it into a chat with a third party.

🧭 Step 3 — ask one question, not a project

The workspace runs one capped task per question. That is a deliberate limit, not an unfinished feature:

• A task has a maximum budget; the spend is counted per attempt.
• There is no automatic paid retry. If a request fails, the task ends and you decide whether to run it again.
• Conversations are saved locally; a new question is a new task with its own limit.

For your first run, ask something small and checkable: "Explain this error message and give me the command that proves the fix." Watch what the answer actually claims, then verify one claim yourself.

🧭 Step 4 — read the receipt

Open Usage (dsh.tvijo.com/usage). For your own account it shows the provider and model that answered, the token counts, the task limit and the spend. The receipt is the only honest source for "what did this cost" — not a marketing estimate, and not a number someone remembers.

If a cost is genuinely unknown, the workspace says so. It does not replace an unknown with $0.00, because a fake zero is worse than an honest blank: it makes budgets look safe when they are not.

📊 What the free preview is, exactly

New self-service accounts receive a sponsored preview: 5 tasks, 50,000 model tokens and USD 0.05 in total, whichever ends first. The shared preview pool is USD 1 per UTC day. The preview supports the cheap routes only; premium routes need paid credit.

This is a preview, not a pricing promise. It is not revenue, it is not a subscription, and it can end. No card is required and there is no automatic debit.

🚧 When this does NOT work

• You need a guaranteed model or vendor. The workspace routes requests server-side. You pick a route, not a specific vendor, and the receipt tells you what answered.
• You need unattended automation. Each question is one task with a cap; there is no auto-retry loop to leave running overnight.
• You need a signed enterprise agreement before sending data. Stop here and read Permissions and data flow first.
• You want to bring your own API key. Connection modes currently save an explicit pilot preference only — no provider key is collected and the route does not switch.

➡️ Next

• Install the editor extension — the same account, inside VS Code.
• Costs, budgets and limits — what the numbers in the receipt mean.
• Recovery after a failed run — what to do when a task ends badly.