Your data, in plain language.

• Account details. Your name, email address, sign-in information, and the workspaces and roles you belong to.
• Code you choose to index. When you index a repository, eligible file contents are sent to ContextStream and stored to build and run your search index. This is true whether your editor uses the hosted connection or the local server.
• Project knowledge. Decisions, lessons, plans, documents, tickets, capsules, and media that you or your connected assistant save.
• Conversation transcripts, when capture is on for an editor (see Your controls).
• Connected tools. Content from the services you choose to connect, such as a code host or chat workspace, and the access tokens that let us reach them.
• Billing details. Payments are processed by Stripe. We receive your plan and payment status, not your full card number.
• Usage information. How the product and website are used, so we can operate, secure, and improve them.

We use this information to run ContextStream for you: indexing and search, memory, agents, sharing, and support. We also use it to keep the service secure and prevent abuse, process payments, send account and service emails, and understand which parts of the product are working so we can improve them.

We never sell your project data. Optional deep project learning uses context-use signals and feedback to improve guidance only for your account, after you explicitly agree.

Some features need AI models, so we send content to AI providers to deliver them. Our search-processing providers handle the content you index so it can be searched. Language-model providers receive the context a request needs when a feature or managed agent generates an answer.

If you connect your own model provider or API key, requests made through it are also governed by that provider’s terms.

We don’t use your project data to train AI models, ours or anyone else’s.

Deep project learning starts off. If you turn it on, ContextStream records which guidance was shown, followed, ignored, helpful, or corrected. It uses those signals to improve your account’s guidance and suggest personal lessons, preferences, playbooks, and project guidance.

Your learned history and generated guidance belong to your account. They do not change teammates’ guidance. We do not sell learned data or use it to train AI models. Service-managed AI providers used for this feature are covered by no-training terms; a model provider you connect yourself remains governed by your agreement with it.

Review suggestions, forget guidance, and revert applied suggestions on the Learning page. Turn learning off in Account → Privacy or with contextstream-mcp configure --account-learning off. Withdrawal stops collection immediately and queues deletion of your learned history and generated learning items within 30 days. Ordinary content you saved remains yours.

When you coach an agent with the Agent Coach Kit, we store what you coach in your ContextStream account: each workflow and its goal history, each job and its objective, the drafts you are shown (text, links and media), your feedback in your own words with the agent’s reading of it kept separately, the guidance that comes from your feedback and its history, any coaching-effort preference you state, and your verdicts. Coaching workflows are personal: only you can see them.

We use this data only to coach your agents on that workflow. It stays out of ContextStream’s general memory, recall and learning, and it is never used to train AI models. The agent you coach, such as ChatGPT or Codex, reads it while it works on your job; that use is governed by your agreement with that provider.

Connecting the kit by signing in asks only for coaching access: reading and saving your coaching. If you use an API key instead, that key has your account’s usual access, so create a separate key for the kit and revoke it when you no longer need it. Permission to coach is not permission to publish, share, contact anyone or spend money for you.

You can export a workflow at any time, as Markdown or JSON. Deleting a workflow hides it at once and erases it, with everything in it, from our live systems within 30 days. Deleting your account erases all of it. Links to coaching media expire within 15 minutes.

When you approve the kit’s connection, we keep a small record of the approval: which app you approved (such as ChatGPT or Codex), that it was for the coaching connector, and when, and whether your account was created through that connection. It holds no sign-in codes or tokens and none of your coaching, and it is deleted with your account. We use it, and whether an account later connects another source such as an integration or an indexed project, to count how many accounts the kit brings; those counts are totals.

A workspace is the boundary for your data. Personal records are visible only to you. Project and workspace records are visible to members of that workspace, according to their role. On Team plans, conversation transcripts are personal by default.

A capsule shares only the context you select, with the audience you choose, and you can revoke it. ContextStream doesn’t widen what your connected assistant is allowed to do.

We may disclose information when the law requires it, or to protect the rights, safety, and security of our users and the service.

We rely on service providers to run ContextStream, including cloud hosting (OVHcloud), network delivery, object storage and model routing (Cloudflare), AI processing (see above), payments (Stripe), email delivery (Resend), and product analytics (PostHog). Email us for the current list.

We use product analytics through our own domain to see how ContextStream is used. Session recording is off, and IP addresses aren’t stored.

Our website also uses advertising measurement tags, such as those from Google, LinkedIn, Microsoft, Reddit, TikTok, and X, to understand which campaigns bring people to ContextStream. You can limit these with your browser’s privacy settings or a content blocker.

• Choose deep project learning. It requires a separate, unchecked consent checkbox. Setup defaults and agents cannot turn it on for you. You can withdraw at any time in Account → Privacy.
• Choose what’s indexed. Git ignore rules and a .contextignore file are applied before upload, and secret-looking files, environment files, keys, and certificates are skipped. If a repository must never leave your machine, don’t index it.
• Turn transcripts off. Run contextstream-mcp configure --transcripts off and --hook-transcripts off to stop capture for connected editors.
• Remove what you’ve saved. Documents, decisions, sessions, projects, and individual indexed files each have their own delete controls.
• Export. Saved decisions, memory, and knowledge can be exported as JSON, CSV, or Markdown through the API.

Read the full guide to indexing and capture controls.

Deep project learning events and activity signals are kept for up to 180 days. Usefulness summaries remain while learning is enabled. After withdrawal, learned history and generated learning items are deleted from live systems within 30 days; consent audit records are retained to document your choice.

Coaching data from the Agent Coach Kit is kept until you delete it. A deleted coaching workflow is hidden at once and erased from live systems within 30 days.

We keep your data while your account is active. Deleting a workspace removes its contents from our live systems. Deleted data can remain in backups until those backups expire on their regular rotation.

To delete your account entirely, email [email protected] and we’ll take care of it.

Traffic between your tools and ContextStream is encrypted with TLS, and access tokens for connected services are encrypted before we store them. Read how we protect your data.

Depending on where you live, you may have the right to access, correct, export, or delete your personal information, and to object to or restrict how we use it. Email [email protected] and we’ll respond within the time the law requires.

ContextStream is built for professional use and isn’t intended for children under 16.

When we make a material change to this policy, we’ll update the date at the top and let you know.

Privacy questions: [email protected]. Security reports: [email protected].