AI accounts and models

Signing in to Neural Labs gives you access to the desktop. Connecting an AI account lets Alshival run requests. These are separate steps, even if you are the only user and the administrator.

Connect your personal account

1. Open Settings → Model Provider, then the OpenAI card.
2. Under Codex sign-in for Terminal, choose whether to Keep sign-ins separate (the default) or Use the same Codex sign-in. Then choose Connect OpenAI. The browser opens the OpenAI device sign-in page in a new tab; the one-time code appears in the card when the native login starts.
3. Enter the code on that page and finish sign-in. Keep Settings open while the card checks your account, loads models, and selects the default available model for your chats.
4. Send a request in a private Alshival conversation to verify the account.

If the device-code attempt ends or expires, the card stops waiting and offers Connect OpenAI again for a fresh code.

If OpenAI opens but the card never shows a code, the installed runtime must support the pinned Codex CLI's numbered device-login prompt. Upgrade that instance through its supported operator update process; changing portal routes alone does not update the native login adapter. The code stays private to the person who started sign-in.

Additional personal, shared, Team, and background connections can be managed under Advanced connections. These require explicit owner and model choices.

ChatGPT subscriptions and OpenAI Platform API usage are separate. The OpenAI card uses native ChatGPT sign-in and does not switch to API billing on failure. If OpenAI asks you to enable device-code login, follow its account security instructions and start a new sign-in. Reload models refreshes the catalog; it does not resume a paused connection. Pause and Resume are under Advanced connections.

Share Codex sign-in with your terminal

The personal OpenAI card offers Codex sign-in for Terminal before or after connecting. Choose Use the same Codex sign-in to let Neura and your private workspace terminal use the same Codex credentials. A login, token refresh, or logout in either is reflected in both. The shared Codex directory also contains its configuration and CLI session state. Your other terminal files stay in place.

Open a new private terminal after changing the option. Choosing Keep sign-ins separate restores the terminal's independent Codex directory without deleting either sign-in. Existing shared terminals close on their next authorization check when sharing is disabled; existing separate terminals keep their current mode until reopened. You can check the new terminal with codex login status.

This option applies to your own personal ChatGPT connection and the workspace's private Terminal. It does not change the physical host's staff accounts, other members, or Team terminals. Team terminals have a separate home. Claude, shared, Team and background connections do not offer this option.

Connect Claude

Open Settings → Model Provider → Anthropic, then Connect Anthropic. This opens Anthropic in a new tab. Complete sign-in, paste the returned code into the card, and choose Connect. Models load automatically and one tiny Claude request verifies access before the card reports readiness. No Terminal is needed. See Connect Anthropic for reconnect and retry guidance.

This flow needs no additional public DNS, OAuth callback registration, or inbound port. The CLI and browser need outbound access to Anthropic. Claude keeps the subscription credentials in its own persistent account directory.

You can keep ChatGPT and Claude connected together. Choose the provider and model in Agent defaults, then save. A conversation's explicit model selection can choose either connected provider. Pausing or disconnecting one provider affects requests using that provider and retains your chats and saved defaults. Reconnect uses a fresh native login home; an old Claude native session may need a new conversation after the account is replaced.

Which account runs a request?

Work	Account used
Private Alshival conversation or a personal skill test	Your personal account for the selected model provider
Manually running an AI task automation from the desktop	The person pressing Run's personal account
Scheduled automation	The automation's assigned agent; main uses the background account
Team Chat	The explicitly selected native Team connection, including requests from members without a personal connection
Alshival realtime voice and voice-memo transcription	The server's OPENAI_API_KEY, configured separately for audio
codex run directly in a terminal	Your personal Codex sign-in if sharing is enabled; otherwise the terminal's separate login cache

A missing, paused, or unusable connection does not silently fall back to another person's account, the other personal billing method, or the shared audio API key. Scheduled and manual automation runs can overlap; the Automations guide explains run behavior.

Connect background and Team accounts

Administrators open Settings → Workspace to configure these separately:

• Background ChatGPT connection supplies the main agent used by scheduled AI work and background tasks. Connect it if you plan to use those features.
• Background AI Claude connection provides an independent Claude sign-in or workspace API key. Saving an API key explicitly selects separately billed API usage for this workload. It does not change your model default.
• Team Alshival uses a native Team connection. Under Advanced connections, connect it, load models, choose one, and select Use for Team Chat. Team Chat requires that saved connection; a disconnect leaves Team turns unavailable.
• Voice controls supported audio models and voice selection. Supply the audio API key through the protected deployment configuration and apply it with the normal operator update process.

Connecting a background or Team account does not connect an administrator's personal account. Use an account authorized for the intended workspace use.

Choose model defaults

Model Provider offers supported models and reasoning levels for your account. Use follow-latest to track the supported recommendation for that workload, or pin a model when you want a fixed choice. Save the selection and check that it has applied; a pending or failed change is not an active model configuration.

A private conversation can override its agent defaults. Clearing the conversation's override returns to the agent default. Automations may have explicit model settings of their own. Changing your personal default does not rewrite scheduled jobs or Team defaults.

Claude models use the native Claude runtime. The model list reports account readiness and runtime metadata; Anthropic can still reject a model because of plan eligibility or usage limits. A failed request does not switch billing or providers automatically. Arbitrary additional provider connections remain outside the fixed OpenAI and Claude bindings.

Credential storage and recovery

OpenClaw keeps personal ChatGPT and OpenAI API-key credentials in separate native profiles in the persistent workspace volumes. A key travels through the protected Settings route and workspace control channel and is never returned to the browser. The browser receives connection status and the temporary ChatGPT login code, not OAuth tokens. Workspace Claude API keys stay in native credential profiles; subscription logins stay in the owning Claude configuration directory. All approved developers share the workspace operating-system trust boundary; see Sharing and privacy.

The separate terminal CLI can be connected with bin/neural-labs workspace codex-login when needed. It does not configure Alshival. For personal, background, or Team sign-in problems, start with Troubleshooting.

Live voice with your selected agent

In a private Alshival conversation, press the wave button with an empty composer to start live voice. OpenAI handles speech while your selected agent answers requests and uses tools. Your conversation's model override and approval policy continue to apply. Approve pending actions in the chat UI; voice cannot approve on your behalf. Missing or paused provider connections do not fall back to the audio key.

In a shared channel, Talk to Alshival starts a local audio call with the channel's configured Team agent. Transcripts and results are shared with channel members. The existing wave control still records voice memos.

Mute or end the call using its controls. Calls last up to five minutes and may be restarted. Ending voice stops microphone capture without discarding accepted agent work. Request transcripts and task replies remain in normal history; spoken renderings and audio usage are recorded as session events, without raw call recordings. If a connection fails after an action starts, inspect the chat before retrying.

The deployment's existing server OPENAI_API_KEY funds audio independently of personal provider credentials. Configure it through the protected deployment configuration. A managed installation can enforce an external admission policy; self-hosted installations do not require an Alshival commercial plan.