Which Claude sign-in method to pick
Claude Code login is usually the Claude sign-in method to pick; choose Sign in to Claude, your API key, CLI proxy or Managed accounts only when you need them. Whichever you pick applies to every Claude session, and Managed accounts only appears when it is available to you. This is Anthropic-only — every other provider keeps its own separate login and is unaffected.
On this page
What it is
Omniscio gives you a choice of how your Claude (Anthropic) sessions authenticate — a single sign-in method that applies to every Claude session you run. For most people the answer is Claude Code login, the default; the others are there when you want something different. You pick in Settings → Accounts → the Claude tab, and there are up to five methods:
- Claude Code login (the default) — you sign in with your own Claude account by running
claude auth login, and Omniscio simply lets that login (~/.claude) power your sessions. Omniscio injects nothing and stores and tracks nothing about that credential, and you never have to touch it again. You don't have to type the command either: the card's Sign in with your Claude account button opens that terminal for you — the same thing first-run setup does when you choose "Sign in with your Claude account" there. That button is Windows and macOS only; on Linux Omniscio tells you to run the command yourself, because an interactive sign-in needs a terminal. Once you finish the sign-in in your browser, Omniscio notices by itself within a couple of seconds: the card changes to Signed in to Claude and the account badge in the header updates too, whichever screen opened the sign-in — including the Claude login row in Connected Tools — and with no restart. Check again looks straight away. With this method the account widget shows one Claude Code row without usage bars, because the per-account usage list belongs to Managed accounts. That row carries your email and plan. On a Mac, where Claude Code keeps its login in the Keychain rather than in a file, Omniscio asks Claude Code itself who is signed in (claude auth status), so you still see them with no Keychain password prompt; if Claude Code cannot answer, the row reads simply "Claude Code CLI". - Sign in to Claude — you sign one Claude account into Omniscio itself, and your sessions run on it directly. One account, yours alone, nothing pooled or shared. Of the public methods this is the only one whose credential lives in Omniscio's own account list.
- Your API key — you paste your own Anthropic API key into a dedicated, isolated slot, and your Claude sessions run directly on Anthropic's API with that one key. No login, no pool, no switching — Omniscio never treats it as an account. You get a key at console.anthropic.com.
- CLI proxy — your sessions run through your own CLIProxyAPI, the open-source proxy where you sign your AI accounts in once and point every tool at it (T3 Code and others plug into it the same way). Omniscio stores only the proxy's address and one of its client keys — never the accounts behind it — and sends every Claude session, and Omniscio's own small AI helpers (session titles, summaries, suggestions), to that address with that key. The proxy decides which of your accounts serves each request, so any account type it holds — Claude, Gemini, GPT/Codex, Grok, Qwen and the rest — can do the work.
- Managed accounts — Omniscio holds one or more Anthropic logins and/or API keys as a pool and picks the freshest one for each session (add, switch, load-balance, auto-recover from rate limits). This is the multi-account experience described in add-a-claude-account.md and account-pool.md. It only appears when it's available to you — a fresh install that has never added an account won't see it (see below).
Which one to pick? If you already use Claude Code on this computer, Claude Code login is the least surprising choice — it is the default, and Omniscio keeps its hands off your login. Pick Sign in to Claude if you would rather sign in inside Omniscio and leave your terminal setup alone, Your API key if you bill Anthropic per token instead of using an included subscription, or CLI proxy if you already keep your accounts in CLIProxyAPI and want Omniscio to use them too. Managed accounts is the one to choose when you want more than one Claude account working together.
One caveat on Claude Code login: a claude.ai Free plan can sign in but cannot run Claude Code, so a Free login counts as an account and not as a working credential. If that is your plan, Claude Code login will not start your sessions.
The methods are completely separate. Whichever one you pick is the only thing that credentials your Claude sessions — the others are never touched. In particular, "Your API key" and "Claude Code login" never reach into the managed pool, and the managed pool never leaks into them.
This is Anthropic-only. The sign-in method changes how your Claude sessions authenticate and nothing else. Every other provider — Codex, Gemini, and the rest — keeps its own separate login or key and is completely unaffected no matter which method you choose.
Where to find it
How to use it
Open Settings → Accounts and go to the Claude tab. At the top you'll see the sign-in method picker — a small segmented control with the methods available to you. Claude Code login, Sign in to Claude, Your API key and CLI proxy are always there; Managed accounts is shown first only if it's available to you (see the next section).
Pick the method you want. Because switching your sign-in method changes how every Claude session authenticates, Omniscio shows a "switch your sign-in method?" confirmation first. Confirm it and the change applies to all your Claude sessions going forward; cancel and nothing changes.
Fill in what that method needs. The area below the picker shows exactly the one thing the selected method requires:
- Claude Code login → a card with a Sign in with your Claude account button, which opens
claude auth loginin a terminal for you (there's a Check again button for once you've finished in the browser). There's nothing to paste — Omniscio uses the login the CLI writes to~/.claude. On Linux the button is unsupported, so runclaude auth loginyourself. - Sign in to Claude → its own card with a Sign in with your Claude account button, which opens Omniscio's own Anthropic sign-in in your browser and stores that single account for you. It is a separate sign-in from the
claude auth loginone above. Sign out removes it again. - Your API key → a password-style box to paste your
sk-ant-...key (it's stored encrypted on your machine, never shown back to you, and never sent to the browser). Get one at console.anthropic.com. - CLI proxy → a card with the proxy's address (already filled in with the usual local one,
http://127.0.0.1:8317), a password-style box for one of theapi-keysfrom your CLIProxyAPI config file (stored encrypted, never shown back), and a Test connection button. The test reports how many models your proxy offers in total and how many are Claude models — or says plainly that it lists none, rejected the key, or couldn't be reached. Each successful test also saves the proxy's model list: every model appears in the session model picker under Your CLI proxy, so you can start a session on a Gemini, Qwen or any other model your proxy serves, and that session keeps every one of its calls on that exact model. Below the test you'll find Model for AI features (which proxy model Omniscio's small AI helpers use — Automatic picks a fast Claude model if your proxy has one, otherwise the first model it lists) and two switches, Run GPT sessions through your proxy and Run Grok sessions through your proxy, both off by default; turning one on sends that provider's sessions to your proxy instead of Omniscio's built-in one. Work that goes through your proxy is recorded at $0 (the token counts are kept) and never counts against Omniscio's daily dollar limits, because your proxy's own accounts pay for it. A plainhttpaddress on another computer gets a one-line warning, because the key would travel unencrypted. If the address or key is missing when a session starts, the session stops with "finish setting up your CLI proxy" instead of quietly using another sign-in. Cloud and SSH sessions keep their own sign-in; this method feeds sessions on this computer. - Managed accounts → nothing extra here; your account rows live in the rest of the Accounts page (add-a-claude-account.md).
- Claude Code login → a card with a Sign in with your Claude account button, which opens
That's it — it applies everywhere. There's no per-session choice. The method you pick credentials every new Claude session you start.
How it behaves
Add a key from the setup banner
On desktop, the setup banner's Add API key button opens a popup without leaving your current screen. Paste your Anthropic API key and choose Save. The popup checks the key using the existing provider validation, stores it securely, and closes after saving succeeds. If saving fails, the popup stays open with your entry and an error message so you can try again. Escape or the close button cancels while no check or save is running.
Saving preserves your selected sign-in method. With Managed accounts, the key is added to that account list. With any other method, it goes into the isolated Your API key slot. If you currently use Claude Code login, Sign in to Claude or CLI proxy, choose Your API key in Settings when you want Claude sessions to use the saved key.
Go to Settings opens Accounts on the Claude tab and reveals the key form for Managed accounts or Your API key. For the other methods, it reveals the sign-in method selector because their key form is hidden. This works even if you last viewed Codex or Grok. Browser and phone users continue to see guidance to connect from the desktop app.
Who pays for Claude
Settings → Accounts opens on a "Who pays" section, and its first card is Claude. It says in one line who pays — for example "Paid by your Claude subscription (Claude Code login)" — and shows two rows:
- Your sign-in method, with how it is billed: Included in your plan for Claude Code login, Sign in to Claude and Managed accounts; Billed to you per use for your own API key. You change the method on the Claude tab further down the page.
- Omniscio credits, an on/off backup row. While it is on, your credits pay for a Claude session only when your sign-in method has no working credential at all (for example you never ran
claude auth login). It is never offered beside your own API key, which never falls back to credits. A brand-new install starts with it on; an existing install keeps exactly what it could do before the update.
When your Claude plan hits its limit, the card lets you choose (only while the credits row is on): Wait for it to reset — the default, exactly what happened before — or Use Omniscio credits, which moves that session onto your credits once, tells you so in the chat, and goes back to your plan when the limit resets. An account that is simply busy or maxed out is a limit, not a missing sign-in, so it never spends credits unless you picked Use Omniscio credits.
When a session does run on your credits because no working Claude sign-in was found, it says so once in that chat — "This session is running on your Omniscio credits because no working Claude sign-in was found" — with an Open Accounts link to sign in. It is never repeated within the same session. If your accounts are all broken (signed out, shut down for billing or unreadable) rather than simply busy, that is the message you see, never "Your Claude plan hit its limit".
Every Claude session also shows Paid by: … on its model badge — your Claude subscription, your Anthropic API key, or Omniscio credits.
When a sign-in stops working
- Re-authenticate buttons do the right sign-in for your method. The button on a session's sign-in error, the startup "login token expired" notice and the Council panel all start the sign-in your current method uses: Claude Code login opens its own sign-in window (finish it there, then re-send your message), Sign in to Claude and Managed accounts open Omniscio's Anthropic sign-in for that account, and Your API key or CLI proxy opens that method's settings form.
- A free Claude plan is called out. If the account behind your Claude Code login is on Claude's free plan, its card says "Free Claude plan — sessions can't run": Claude Code needs a paid plan (Pro or Max). Switch to a paid account, or use your own API key.
- Sign in to Claude shows a broken login and fixes it in place. If your signed-in account stops working, its card says "Your Claude sign-in stopped working" with a Sign in again button. Signing in again with the same account repairs it where it is — nothing is removed first. If you sign in as a different account instead, it is added beside the old one and your sessions run on the new one; each account has its own Sign out.
- A login your organisation blocks says so. When Anthropic refuses an account because its organisation doesn't allow this sign-in, the account row says "This organization doesn't allow this sign-in here — sign in with a different account" instead of a bare "Authentication failed".
- After four failed retries the message says what happened — that your sign-in couldn't be renewed, or that Claude kept refusing it even after renewing — and offers Re-authenticate.
How it works
Your choice is a single setting that both the UI and the session-spawn path read through the same resolver, so what the Settings tab shows is exactly how your sessions will be signed in — they can never disagree. When you choose Your API key, that key is injected as the session's ANTHROPIC_API_KEY from its own isolated slot and never mixes with any account. When you choose Claude Code login, Omniscio injects no credential at all and the claude CLI falls back to its own ~/.claude login. When you choose CLI proxy, each session is pointed at your proxy's address with your proxy key as its only credential, and side runs that branch off a session use the same proxy. If the proxy rejects the key, or has no signed-in account that can serve the model, the session stops with a message naming the proxy rather than retrying. Agents cannot change the proxy address or key through Omniscio's control API. When you choose Managed accounts, sessions use the account pool exactly as before.
Why "Managed accounts" might not appear. Managing multiple Anthropic logins as a shared pool is an owner/internal capability, so it's hidden by default. It shows up only when it's genuinely available to you — most commonly because your install already holds pooled accounts (so nothing you were using disappears), because your install ran the pool before and Omniscio recorded that (the record outlives the account list, so a reinstall that keeps your user folder does not take the option away), or because an owner has turned it on for you. That record is made on this computer the first time the pool is used and the first time you pick Managed accounts yourself, and it is saved outside the app's data folder so rebuilding or reinstalling the app cannot erase it. If your account list was wiped before any version that writes that record ever ran, an owner can approve Managed accounts for your Omniscio account from the admin dashboard's per-user page (the Managed accounts (pool) checkbox), and after you update and restart the option comes back on its own — Omniscio reads that approval from your account each time it starts, so there is no file to edit, nothing to re-enter, and no need to sign out and back in. On a brand-new install with neither a record nor that approval, you'll see just the four public methods, and Omniscio defaults you to Claude Code login rather than the pool. When you're on a public method (Claude Code login, Sign in to Claude, Your API key or CLI proxy), the managed account rows and the "Log In with Anthropic" button are hidden entirely — you see exactly one Claude credential entry point, that method's own control — while the always-available "Other AI tools" list and its toggle stay put. Only the Anthropic credential is governed this way; every other provider is untouched.
How Omniscio knows you're set up. Every method counts as being signed in — the account badge in the header, and the "you haven't set up an account yet" banner and startup reminder, each recognise whichever method you're on. That matters most right after first-run setup: Your API key and Claude Code login deliberately store nothing in the account list, so before this they were mistaken for "no account at all" and you'd be asked to sign in again despite having just done it. Omniscio also re-checks your credential the moment setup closes, so a claude auth login you completed in the terminal mid-setup is picked up straight away rather than only on the next restart. If you skipped sign-in entirely, or picked Your API key without pasting one, the reminder still appears — it's telling you something true. A Claude Code login is read as signed in even at the moment Claude Code is rewriting its own credentials file (it does that every time it renews your token): a file that is there but momentarily unreadable counts as your login, so a session started in that instant uses your own account instead of being moved onto Omniscio credit, and you are never asked to sign in again because of it.
A test copy of Omniscio never signs in to your real Claude Code login. A second copy started for testing (a developer's dev copy, an automated test run, a sandbox) shares your computer but not your login: its Sign in with your Claude account button, and the Claude login row in Connected Tools, refuse with a message instead of opening the terminal, because that sign-in would replace the Claude Code login your main Omniscio uses. On the Accounts card that refusal is shown to you as written, naming both ways forward, rather than a generic "something went wrong" — and the Claude login row in Connected Tools, and the tool list in first-run setup, show it as written too. To sign a test copy in, start it with its own CLAUDE_CONFIG_DIR folder; the sign-in then writes into that folder, which is the same one that copy's sessions read. Your main Omniscio is unchanged.
When the key won't save or the login fails
Onboarding's Get help buttons on the sign-in and API-key steps come here, so this section covers the failures those steps can hand you.
- "Couldn't add the API key. Please try again." — the save did not complete, so the previous setting is still in place. Paste the key again; if it repeats, sign out and back in and retry.
- Paste only the key. The box takes the value starting
sk-ant-…, without surrounding quotes, spaces or anANTHROPIC_API_KEY=prefix, which a key copied from a config file often carries. - The key saves, but sessions report an auth error. A key can be revoked, disabled, out of credit, or belong to an organization that has not enabled the model. Check it at console.anthropic.com and replace it in Settings → Accounts → Claude; sessions pick the new key up the next time they start.
- "Sign-in failed — Anthropic could not verify the login. Please try again; if it keeps failing, check status.anthropic.com." — the last step of Sign in to Claude (Omniscio's own Anthropic sign-in) failed for a reason other than the rate limit, which has its own longer message about waiting. Often it is a real Anthropic outage, so check the status page before retrying in a loop.
Related
- add-a-claude-account.md — the "Managed accounts" method in depth: adding Claude.ai logins or an API key to the pool
- switch-active-account.md — within the managed pool, how Omniscio picks and switches the active account per session
- account-pool.md — the tier-list the managed pool uses to choose an account
- remove-a-claude-account.md — taking a managed account back out
- switching-providers.md — using a different provider (Codex, Gemini, …) for a session; those are unaffected by your Claude sign-in method
Last verified 2026-11-12