Providers
Keys are stored on your server and never sent to the browser. Set them as environment
variables instead and they become read-only here.
Appearance
Theme
Applies immediately, and only to this browser.
Language
Applies immediately, and follows your account onto every device.
Emails
Help
Guide
Show the five-step guide for people just starting out.
Profile
The assistant keeps these in mind in every conversation. Saved when you leave the box.
Account
Log out of all devices
Every browser signed in to this account is signed out, this one too.
Sign out of this browser
Two-factor authentication
Active sessions
Security activity
Sign-ins, password and two-factor changes, keys, exports and shared links on this account — never what
was said. Addresses are kept only to their network. Kept for 180 days.
Privacy
What you write stays in this app's own database. It reaches the model you chose only as part of a conversation, and that model's provider handles it under its own terms.
The privacy notice — what is processed, where it goes, and your rights
Usage policy
Preferences
For models reached through OpenRouter. Strict sends requests only to zero-data-retention endpoints
that neither store nor train on what you send. Many free models are not served that way, so they
will stop working while this is on.
Email addresses, phone numbers, ID, tax, insurance and bank-account numbers, card numbers, addresses, birth dates, number plates, IP addresses, keys and passwords, and names written after a label or a title, or a Vietnamese full name of three words or more, are replaced with placeholders before anything is sent, and put back in the reply and before any tool runs. A PDF goes as its text while this is on, and sound and video do not go at all. Pictures and scanned pages go as they are, and a name of one or two words, a foreign name with no title, or anything identifying by context is not caught. The model works with the placeholders, so answers about those details can be less precise.
Conversations not continued for this long are deleted with their files. Pinned conversations are
always kept.
On this deployment, page views and page speed are counted by Vercel Web Analytics and Speed
Insights: no cookies, only the page's address without anything after ? or #, never what you write.
A browser set to Do Not Track or Global Privacy Control is not counted.
Your data
Download everything this account holds as one file, or bring in conversations exported from Claude
or ChatGPT (their conversations.json) or from another Synapsez. Keys and connector tokens are never
exported.
General
Mode and reasoning effort live on the button beside
Send, where they apply the moment you press it.
How the assistant's tools are loaded in new conversations.
Each turn re-sends the whole conversation, so a long one eventually stops fitting. The
ring in the header shows how full it is; you can also fold it up yourself from there.
Visuals
The reason to ask for a report is to read it. It opens once per turn and closing it
means closed, so it never reappears while you are working — and never on a narrow
screen, where the panel would cover the conversation.
Skills
Skills have moved to Customize.
Memory
What the assistant remembers between conversations, and what it may look back at. All of it stays
in this app's database; it reaches the model you chose only as part of a conversation — and that
model's provider handles it under its own terms. When the assistant changes or removes a note, up
to five earlier versions are kept, with where each was saved, so you can bring one back; editing or
deleting the note yourself keeps none. A rule is never written or changed without your yes; an
event stops being used after its date.
Bring memory from another assistant
Ask Claude or ChatGPT to "write out everything you remember about me", paste the answer here, and it
becomes one note you can read and edit below.
Saved notes
System
Synapsez version
Keyboard shortcuts
Every shortcut the app answers to.
Browser use
Computer use
Its own computer in the cloud: it runs code there, makes files and installs what it needs. Never your own computer.
Address safety
Is this site's address safe from spoofing?
Checked live: signed DNS answers (DNSSEC), HTTPS that browsers insist on (HSTS and the preload list), who may issue certificates for it (CAA), and how well the address it sends email from is protected.
Tools from outside Synapsez — Figma, Jira, a database — through MCP servers. Every tool from one asks before it runs.
Advanced settings
Extension settings
Where servers that are programs run
Extension developer
Developer tools warning
These are for adding servers yourself. A server added by hand runs a program or reaches an address you name — only add what you trust. Every tool from it still asks before it runs.
MCP servers you add yourself
MCP servers add tools from outside this app — Figma, Jira, Sentry, a database, hundreds of
others. Plug one in and the assistant can use it. Nothing here is chosen by the assistant:
you type the command, and every tool from a server stops and asks before it runs.
Need an API key?
Every provider's keys — and the order they are tried in — are under Providers.
Upload a skill
Add a skill to your workspace. A security scan runs when you upload.
Skill file
- .md file must contain skill name and description formatted in YAML
- .zip or .skill file must include a SKILL.md file
Security scan
Runs when you upload
Choose a file to continue.
Families of the assistant's built-in tools. One that is off is not offered to the model at all.
People
Administrators
The owner decides who else helps run this deployment, and what each of them may do.
Themes
Choose the badge each theme carries, and give it to or take it from someone by hand. The top three of a period get the themes carrying that period, or the ones marked for any period.
Billing
What billing took in, what the paid services cost, how full they are, and the money nobody has claimed yet.
Security activity
The newest events across every account on this deployment.