Connect Claude Code
3 minute read · Chats & Local Agents
Connect Claude Code once, then choose it in a project conversation. Your Vortex sign-in and provider sign-in are separate accounts.
- Connect the account
- What to expect from Claude Code
- Verify the account with project work
- Choose settings the provider supports
- When sign-in or a run fails
Connect the account
- Open Settings → Agents & chat and select Claude Code.
- Check installation status. If needed, choose Install and follow preparing, installing and verification progress. An error or cancelled installation needs attention before a run can start.
- Choose Sign in. Complete the supported browser flow and any displayed provider code, then return to Vortex to verify the account status.
- Start a new chat, choose Claude Code and an available model, then ask it to explain the first step of your app idea.
What to expect from Claude Code
Vortex can use the available packaged Claude runtime; a globally installed executable is not universally required. Preserve an existing detected installation and sign-in. Claude can report five-hour and weekly account usage windows when supplied by its protocol.
If the app reports unknown or signed out, open the connection controls again. An unknown state is not a confirmed sign-in. A provider account rejection needs an eligible account or another supported agent; Vortex does not change provider eligibility or include its tokens.
Continue with your first app, model and usage choices and permissions.
Verify the account with project work
Create a fresh project conversation, choose Claude Code and ask it to inspect the app before editing: “Explain what starts the frontend and API, then suggest one small improvement.” Confirm that the result describes this checkout. When you are ready, request the improvement, answer any approval in that exact run and inspect the changed files.
Claude can use a packaged runtime path. A missing globally installed claude command does not by itself prove that the Vortex integration cannot run. Use the installation and connection state shown by Vortex; preserve an existing detected system installation and provider sign-in when troubleshooting.
Choose settings the provider supports
Select a discovered or supported custom model through the composer. Thinking controls and access choices depend on the model and protocol. Use Supervised when learning the workflow, and read the scope of an approval before granting it. A permissive access mode does not turn a request to edit local code into permission to publish or send external messages.
Account usage can include five-hour and weekly windows. Context consumption is a different measure, and old quota samples should not be read as current after a reset. Your Anthropic subscription and Claude tokens remain separate from Vortex.
When sign-in or a run fails
Complete the supported browser/account flow, return to the connection dialog and wait for verification. If the account remains unknown, treat that as an unresolved check, not a confirmed connection. For a failed run, retain its error and check model/account availability before retrying. Stopping a run invalidates its pending approvals; approve a new request only in the new run where it appears. Permissions and steering cover those interactions.