Organization and members
Each person has one organization of their own. This page shows how to sign up and sign in, and how to set the organization's limits.
Click "Organization" in the navigation. The settings have five tabs: "Policy", "Usage", "Billing", "Audit log" and "My CLI tokens". Usage and billing are on Billing and usage.
Your organization
You get your organization the first time you sign in with Yi Labs, and you are its owner.
It is the only one you have: there is no organization picker, and you can't create another one.
You can't join someone else's organization either: invitations aren't used with Yi Labs sign-in (see Members and invitations).
Connected products such as Yi Guard can ask you to choose an organization when signing in. "Create your own organization" there creates and selects it for that product; it does not change your current Yi Eval organization. Yi Guard opens its own default project for an organization's first owner sign-in, and other members start with read-only Yi Guard access.
Roles
You are the owner of your organization, so you can do everything on these pages: projects, tests, runs, secrets, connections, API tokens, automatic runs, the policy, what the AI does by itself, the audit log and billing.
A change that confirms, launches or spends something is made by a person signed in in the browser, or by the platform within what a person switched on. Auto-approve, for example, confirms drafts; it is set per project and on by default. The AI starts, re-runs, retests or resizes runs only as far as the owner allowed under "What the AI does by itself" (off by default), within its caps, and each of those actions goes in the audit log. A CLI token or the in-app assistant can read and draft, but can't confirm or start tests.
Members and invitations
Each person has one organization, so there is no "Members" tab, and invitations aren't used with Yi Labs sign-in. An invitation link from before doesn't work. Anyone else who wants to use Yi Eval signs in with their own Yi Labs account and gets their own organization.
Sign up
New to Yi Labs? You sign up on the Yi Labs account's own page.
- On the sign-in page, click "Sign in with Yi Labs".
- On the Yi Labs page, sign up with GitHub or Google; these are the hosted account's available sign-in methods.
- You come straight back, signed in, with your own organization.
Connected products use your Yi Labs account; check Account help for the current wallet connections. Tests are paid from your Yi Labs wallet; top-ups and plans are in the Yi Labs account (see Billing and usage).
Sign in
On the sign-in page, click "Sign in with Yi Labs" and sign in on the Yi Labs page the way you signed up. You come straight back. There is no email form or GitHub or Google button here: those are on the Yi Labs page.
If your Yi Labs account is suspended or deleted, you are signed out here too.
The CLI signs in through the browser: see sign in from the CLI.
Organization policy
The "Policy" tab limits the organization's sandboxes. Owners change it; everyone can read it. Leave a field empty for the default, then click "Save".
| Field | Default | Range |
|---|---|---|
| "Sandboxes running at once" | 2 | 0 to 100, and no more than the plan allows |
| "Longest run (minutes)" | 120 | 1 to 1,440, and no more than the plan allows |
| "Sandbox hours per month" | The plan's allowance | Up to the plan's allowance |
| "Evidence retention (days)" | 90 | 1 to 3,650 |
A value above the plan's cap is refused, naming the cap. What each limit does, and the caps of each plan, are in Sandboxes and limits.
Retention period
Every night, the evidence files and sandbox logs of runs older than the retention period are deleted, and so are the traces in Logs (their spans and stored payloads). The report, its verdict and the problems found stay, and so do experiments and their scores. A report whose evidence has gone says so, with the retention period it went after.
Default model
"Default model" on the "Policy" tab is the model the organization's projects use when they follow the organization ("Follow the organization default (…)" in a project's header), and the one the assistant uses outside a project. It offers the same models as a project. Only owners change it; with none chosen, the platform's default applies (Gemini 4 Argon; GPT-6.1 Sol before 24 October 2026). See Models.
What the AI does by itself
"What the AI does by itself" on the "Policy" tab says what the platform's AI may do without anyone clicking. It comes down to four switches, which an owner sets:
| Switch | What it decides |
|---|---|
| "Ask me before the AI changes things" | On (the default): the platform's AI changes nothing in a project by itself — no class switch, no change to the test connection, no hand-over of someone's automations; each waits for a click. Off: when the platform recommends a bigger class, the AI switches the project to it (switch back under "Settings → Sandbox"); when someone who switched on automations leaves, the AI gives them to an owner; when the connection lacks what the tests need, it hands it to "Let AI finish and check" and, once its check passes, retests only those (with "Re-runs by the platform" on too). Nothing a test checks is loosened, and files you wrote for the connection aren't changed. |
| "Ask me before fixing the test connection" | On (the default): when a pre-run or check fails, an AI diagnosis finds the connection at fault, or AI approval finds a draft needs the connection changed, the platform first asks on the project's Targets page ("The platform would spend about $X and about N min more fixing the test connection. Fix it?") and fixes it only on "Fix it"; "Not now" costs nothing. Off: the platform fixes the test connection by itself when it finds it at fault, billed as usual; a round that changed no file isn't charged. See Let AI finish and check. |
| "Online scoring" | On (the default): each project's online scoring rules ("Automations → Online scoring") score the live traces coming in, paid from the wallet balance (and within a rule's daily budget, if it has one). Off: nothing in the organization is scored; the rules are kept. |
| "Re-runs by the platform" | On: the tests start once a pre-run passes; a run whose sandbox didn't start or lost contact starts again once, and one at the time limit fills in once; undecided cases, and ones that passed and failed in the same run, run once more as a run of their own — the first verdict stays and no pass criteria are loosened. Test set questions aren't re-run. Off by default. A project can still switch it off under Automations (re-runs by the platform). |

The AI acts for the person who last saved the page, through the same plan, billing and limit checks as that person's own click, and everything it does is in the audit log. "Most the runs the AI starts itself may spend a day" and "a month" (USD) count what the AI starts, as billed — apart from "The platform AI's daily limit" on the same page, which the platform sets and which holds all the platform AI spends; at a cap the AI stops and mails the owners. For the automatic runs of someone who left the organization, see when the person leaves.
Audit log
The "Audit log" tab lists every managed action in the organization, newest first: when, who, which action, on what, and from which IP address. It covers people's clicks that change, spend or launch something, and the platform's own actions, such as deleting a sandbox. Filter by "Action", and click "Earlier records" to page back. Only owners can read it. Lines are only ever added, and they are kept while the organization exists.
My CLI tokens
"My CLI tokens" lists the tokens reviewctl login saved on your computers, for this organization: each one's name, when you signed in, when it was last used and when it expires. A token stops working after 90 days unused, or 12 months after sign-in. reviewctl logout revokes it on the platform; if a computer is lost or a token leaks, revoke it here with the bin. Each person sees only their own.
Name and default language
The owner changes the organization's name and its default language under "Organization" → "Policy", above the policy. The default language is used for the platform's emails to the organization and for members who haven't chosen their own.
Language
- The web app: the "Language" switch in the navigation (中文 / EN). It is remembered in the browser and becomes your language for emails too.
- Emails go in the recipient's language; someone who never chose one gets the organization's default language (see above).
- The CLI and test runners:
REVIEWCTL_LANG=enorzh; see Command line. - The API: the
x-reviewctl-localeheader; see API and tokens.
Next steps
- Billing and usage: plans, the usage page and invoices.
- Sandboxes and limits: what the policy limits do.
- Security overview: access control and the audit log.