The basics
What is String Cortex?
A service from StringHub that saves every Claude Code conversation from every laptop in your company to one secure place, and lets the right people read and search it.
Is it made by Anthropic?
No. It is built and run by StringHub, and works with Claude Code through Claude Code’s own plugin and managed-settings features. Claude and Claude Code are trademarks of Anthropic.
Which Claude plans does it work with?
Company-wide rollout uses managed settings, available on Claude for Teams and Enterprise. Individuals on Pro or Max can install the plugin themselves. Claude Code used with a plain API key has no organisation, so it isn’t accepted.
Does it work with other AI tools?
It is built specifically for Claude Code conversations.
What is Ask Cortex?
A chat on top of your team’s Claude Code history. Ask a question in plain words (“how did we fix the Redis lock timeout?”) and get one answer written from the sessions that match, with numbered sources that open the exact messages. Follow-up questions keep the context, and you can @ mention colleagues or # attach projects to ask about their work only.
Which AI model does Ask Cortex use?
The one your company admin picks: Claude, OpenAI or Gemini, with your company’s own API key, so answers are billed to your account with that provider. The admin also sets how much the model reasons and who may ask, and sees what the answers cost.
Setup
What do employees need to do?
Very little. Once your admin pastes the settings, Claude Code installs the plugin on each laptop in the next session (it may ask once to allow the organisation’s settings). Each person then makes their own Claude Code token in String Cortex and pastes it when Claude Code asks, and picks a workspace the first time they work in a new project folder.
How long does rollout take?
Claude Code picks up new managed settings within about an hour. Each person connects with their own token the next time they use it; there is no approval queue for admins.
What does a laptop need?
Claude Code, and Node.js 18 or newer on its PATH, which most developer machines already have.
Someone lost their token, or changed laptops. Now what?
They generate a new one in String Cortex (avatar → Claude Code token) and paste it when Claude Code asks. The old one stops working at once. Nothing changes in the managed settings.
Can we make String Cortex compulsory?
Yes. Tick Make String Cortex compulsory under Company → Claude Code settings. Claude Code then waits until the person has connected and chosen the folder’s workspace; the message they sent goes through right after.
What are workspaces?
Separate areas inside your company, for teams, clients or products. Each project folder belongs to one: Claude Code asks once per new folder, remembers the answer for its subfolders, and the conversation waits on the laptop until then.
Where do I find our Claude organisation id?
On a company laptop in ~/.claude.json under oauthAccount.organizationUuid. Or simply wait: the first laptop that is refused shows up under Refused recently with an “Add this organisation” button. Ids look like xxxxxxxx-xxxx-xxxx-xxxx-xxxxxxxxxxxx.
Privacy and data
How do teams share projects with each other?
With access groups. A company admin creates groups that give all projects of the company, hand-picked projects, or every project with certain tags, and puts employees in one or more of them. Employees can also be given tags: every project they work in gets those tags, so a group like “projects tagged #backend” keeps itself up to date.
Does Ask Cortex send our conversations to an AI provider?
Only if your admin turns it on, and only to the provider you chose with your own key. For each question it sends that question, the last few messages of the chat and the best matching messages the person asking may see (never anyone else’s), with common secrets such as API keys and passwords masked. Search itself never calls an outside AI.
What exactly is stored?
The prompts people type, Claude’s text replies and token usage per response. Tool calls and results, hidden reasoning and sub-agent internals are skipped.
Who can read the conversations?
Company admins see their company’s conversations. Employees see their own, plus the projects their access groups give them (all, chosen, or by tag). StringHub staff run the service but never see any company’s conversations.
What about secrets pasted into Claude?
Common secret formats, such as API keys, tokens, JWTs, private keys, password=… and credentials in URLs, are masked before anything is stored.
What if someone uses a personal Claude account on a work laptop?
Only your company’s Claude organisation ids are accepted. Conversations from any other account are skipped, not stored.
Does it change or delete anything on laptops?
No. It only reads the transcripts Claude Code already writes, and adds nothing to anyone’s projects.
Can we delete data?
Yes. Remove a session or a whole project from the workspace, and choose whether it stays out or may be rebuilt on the next sync.
Reliability and hosting
Does it slow Claude Code down?
No. After each reply a short hook starts a separate background sender and returns at once.
What happens if the server or network is down?
Nothing is lost. Lines wait on the laptop and are sent with the next reply once the server is reachable again.
Can messages be duplicated or missed?
Delivery is at-least-once, and the server ignores what it already has, so there are no duplicates. If a gap appears, the server tells the laptop where to resume.
Can we host it ourselves?
Yes, on the Enterprise plan. The whole stack runs with Docker Compose on your infrastructure, including the embedding model.