Account
Deep project learning
Private guidance that improves from your context use, with explicit consent and reversible suggestions.
Deep project learning helps ContextStream notice which guidance works for you. It starts off. In Account → Privacy, read the consent text and tick the unchecked box before turning it on. Learning is optional; the rest of ContextStream works without it.
What changes after you consent
Your account records whether guidance was shown, followed, ignored, helpful, or corrected. Repeated evidence can give useful guidance more prominence, reduce stale guidance, and suggest personal lessons, preferences, playbooks, or project guidance. New preferences, playbooks, and project guidance wait for your review. Strong duplicate merges and coached lessons can apply automatically; you can revert them on the Learning page.
Your Learning page
Choose a workspace and, optionally, a project. Guidance shows the evidence and current state of each item. Suggestions lets you accept, dismiss, and revert changes. Activity shows the underlying history. A private weekly recap summarizes the previous completed week. Established guidance can be copied into your rules.
Forget stops an item influencing your account's ranking. It leaves the original record available to its authorized readers. Revert removes a generated personal item or restores guidance retired by that suggestion.
Scope and data handling
Learned history belongs to your account. It does not change teammates' guidance or make a restricted project visible. Generated items stay personal unless you choose to share ordinary content yourself. Your learned data is never sold or used to train AI models. Service-managed providers used for this feature are covered by no-training terms; your own connected provider follows your agreement.
Events and activity signals are kept for up to 180 days. Usefulness summaries remain while learning is enabled. Consent audit records document your choice.
Turn it off
Use Account → Privacy or contextstream-mcp configure --account-learning off.
Collection stops immediately. Deletion of learned history and generated learning
items is queued, retried through outages, and completed in live systems within
30 days. Ordinary content you saved remains. Turning learning back on waits for
any unfinished deletion, then starts fresh.
contextstream-mcp setup --yes preserves your current choice. An explicit
--account-learning on opens the privacy page for you to confirm in a signed-in
browser. An assistant or an API key cannot consent for you.
If it doesn't work
If the Learning page cannot load, use Try again and check that a workspace is selected. If the consent wording changed, read the new text and confirm again; the checkbox starts unchecked. A failed withdrawal keeps its retry button available. You can also use Account → Privacy to check the current state.
Next
Read Privacy and data handling for the broader data policy, or open Account → Privacy to manage your choice.