Overview
A summary of what this page has loaded: what needs you, your connectors, memory, usage and agent coordination.
Loading overview…
Agents
Follow your agents, review requests, and keep useful documents.
Ready for useful work
Your agents recall relevant memory and can suggest documents worth keeping. Review those suggestions below. Connection, capture and privacy controls live in Settings.
Agent coordination
Recent agent-reported status. A lease shows how long a report counts as current; it does not prove an agent is connected.
Coordination details and advanced controls
Complete agent reports, shared tasks, technical resource claims and claim-release controls. These can cover different projects using this Brain account.
What I asked my agents
Requests you typed, as your agents registered them, with their progress. Tick Seen once you have read the result. Delete removes an entry for good.
Selected documents
The AI selects documents with lasting value and explains why. Depending on your Memory settings, you approve each exact revision here, or the AI keeps what it selected and you can inspect it here. Files read during work are never collected automatically. Existing documents stay stored; delete one from Memories.
Loading document selections…
Agent chat
Choose Chat beside a connector in Connections. A queued message waits until that agent reads it.
Chat
Usage details
Usage
These are model-agnostic estimates of the incremental Brain traffic visible to the connector: tool names and arguments the model wrote, plus schemas, results, and errors it read. MCP does not expose the provider’s tokenizer, cached-input accounting, or hidden reasoning tokens, so this is not a provider billing statement. A further — tokens (today) came from your capture hook — visible prompts and final responses that the capture hook sent; this extra transfer added no model tokens. Capture-hook transfer (today): not loaded yet. Older imported daily totals make calendar boundaries approximate for part of this range.
not measured Provider bills, cached-input discounts and hidden reasoning tokens are not visible to Brain. Message counts are measured; token figures are estimates.
Settings
Make Brain work for you. Your choices apply only to this account.
Appearance
Choose a theme for this browser.
Chocolate / VanillaUseful by default
Relevant recall, shared agent coordination, and suggestions for lasting memories keep everyday work simple. Documents start with exact-revision review.
Optional capture and request tracking need your consent and a supported, trusted setup. Existing pauses, exclusions and permissions stay in place. No folder is watched or uploaded.
Connect an agent
Download the setup file, give it to your AI, and say “set this up.” Your AI prepares the connection and asks for your approval before the file’s one-time code is redeemed. Connectors appear under Connector keys below; a new row there does not by itself show which setup file made it.
Setup files are not available for this end-to-end encrypted account: the server refuses to issue them, and its hosted Brain connector does not serve encrypted accounts.
Setup files downloaded in this tab
This page cannot see a setup file being used. It reports only what your connector list shows, and that list refreshes about every 30 seconds while this tab is open.
Guided connector and conversation capture setup supports Claude Code and Codex. Other MCP clients can connect manually, but may not support conversation capture. Capture is optional and off until you approve it; it records visible prompts and final responses, plus the optional sources you turn on in Memory settings, and never uploads documents. The setup file is a small JSON document with a one-time code that expires in 15 minutes at the latest; a service restart can end it sooner. It creates a least-privilege online connector you can revoke below; it never creates a local Brain corpus. Setup guide.
Connector keys
Keys your agents use to connect. Revoke any time.
“Last credential use” is the last time the server saw a key or its capture credential, including refused or failed calls. It does not show that an agent is running or that a call succeeded. New key makes a full-access key for manual MCP setup (it can delete memories); a setup file above makes a least-privilege connector instead.
This account is end-to-end encrypted. The hosted Brain connector refuses encrypted accounts, so keys listed here cannot save or recall memories through it, and chat is not available.
Loading your memory settings…
Conversation capture
One switch for every computer. While capture is paused, the server refuses captured turns from all of them and stores nothing new; connector keys keep working. Revoking a connector key in Settings > Connections stops that one computer for good.
Memory collection
Your settings, for this account only. Nothing is collected from before you turn a source on, and other accounts are never affected. Turning a source off stops new collection at once; what was already kept stays until you delete it.
Recommended: let agents propose useful sources and final documents, review each exact revision, and keep only what will help future work. This is already the default. To keep a request checklist, turn on its switch and review the capture setup on your computer.
Connectors allowed to keep documents without review (AI-selected mode):
Advanced document controls
Only documents your AI deliberately selects can be kept. The selected filename and exact revision are recorded; exclusions, pause and connector permissions still apply. Changing a setting here does not install or trust hooks on a computer.
These settings changed in another tab. Your draft is still here. Review the current choices below before deciding what to save.
Correct an old fact
Facts that moved
Record a change with its date. Recall then flags older memories that still describe the old state.
Sources that are not available
Automatic compaction capture and transcript recovery are unavailable. Enabling an authorization cannot make them work. Your existing source settings are preserved. Visible prompts and final replies use the separate capture switch.
Privacy & retention
Account & data
Export your saved memory, reset your password, or review permanent account deletion.
Memories
What your agents have saved. Search, filter, add a memory, or delete one.
Your memory
Search looks at names, descriptions and the first 800 characters of each memory, so a memory can exist without matching. Memories the nightly tidy archived are not listed here; Export includes them.
Search and filters are off: the server refuses plaintext memory requests for this end-to-end encrypted account.
Updating memories…
Loading…
Add memory
Save a thought, preference, or decision for every connected AI.
Importance changes retrieval attention, not whether a memory is true or verified.
Saved as a note marked unverified. Each save creates a new note, so saving the same text twice stores it twice.
Not available for this end-to-end encrypted account: the server refuses plaintext saves for it.
Health
Not loaded yet.
Open questions
Things your memory wants answered. Tap one to look it up.
None right now. When your memory needs an answer, questions appear here.
Showing the newest 50. The server lists no more than 50 open questions and sends no total.
Reflexes
Read recorded lessons
Lessons from recorded mistakes. An agent's session-start briefing (brain_wake) lists them, up to 12 at a time.
None listed. Lessons appear here when an agent records a mistake.
Map
How your memories cluster and link.
Scroll inside the map to reach its zoom controls and instructions. You can also Tab through its controls.
Your memories, clustered and linked. Bigger nodes are recalled more often. The map is redrawn when this page detects a change; it checks about every 15 seconds while the page is visible.
Activity
Events Vanilla Brain recorded: saves, recalls, tool calls, captures, sessions and Night shift. Times are when the server recorded each event, shown in your local time. Agent names are broad client categories (for example claude or codex), not your connector names.
Activity
No activity yet. Recalls and saves show up here.