New model

Either way you will not be asked about this model again. You can always change the model per conversation from the chip in the header.

Getting started with Synapsez

    Choose a model

    About Synapsez

    Synapsez is an AI assistant: chat with the model you choose, work with your files and, only if you connect them, your Gmail, Drive, Calendar and other Google apps, to do what you ask.

    Create a skill

    Everyone on this site will be able to read this skill and use it with their assistant. Keep anything private out of it — a safety scan checks it first.

    Discard this draft?

    Your changes haven’t been saved and will be lost.

    Keyboard shortcuts

    Memory

    Create a project

    This becomes the project's standing instructions — carried into every conversation in it, so nothing is re-explained.

    Add text content

    Context

    Uploaded files

    Edit project details

    The project's standing instructions — carried into every conversation in it.

    Create workflow

    Each line runs in turn, in one conversation, so a step can refer to what the step before it produced. A run that is interrupted resumes where it stopped — and a step that was already running is never repeated, because there is no way to tell whether it had finished.

    Create scheduled task

    Written as you would say it in the chat, because that is exactly what happens: each run is one turn of a new conversation, with the same tools and the same permissions.

    It pauses at anything risky. Nobody is watching a run at 3am, so it stops there until you look.

    It never runs on its own — only when you press Run now.

    Search conversations

    Settings

    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.

    Theme

    Ranking

    Appearance

    Theme

    Applies immediately, and only to this browser.

    Language

    Applies immediately, and follows your account onto every device.

    Emails

    An email when a new event theme comes out: what it asks and how many places there are.

    Help

    Guide

    Show the five-step guide for people just starting out.

    Profile

    Sign in with this or your email. 3–24 letters, digits, dots or underscores, starting with a letter.

    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
    Change password

    Changing it signs out every other device — this one stays in.

    Delete account

    Deletes the account and everything in it — conversations, files, notes, projects, keys and tasks — at once and for good. Export your data first if you want a copy.

    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.

    A smaller model compares what you asked with what is about to happen, and you are asked when they do not match or it cannot tell. Uses your key: a few hundred tokens per action it checks. One more fence, not a guarantee.

    Conversations not continued for this long are deleted with their files. Pinned conversations are always kept.

    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.

    Export my data
    Import conversations
    Shared chats
    Shared artifacts
    Uploaded files
    Your feedback
    Indexed documents
    Memory preferences

    Billing

    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.

    Let the assistant search the connector directory and surface ones relevant to your conversation.

    Before an answer is kept, every web address in it is checked against what was read. One that was not read is corrected, or its link is removed.

    After the assistant reads the web, every number, percentage and date in its answer is looked for in what it read. One that is not there is corrected or marked "(unverified)", including figures it worked out itself. A correction uses one more request.

    How many tools the assistant may use for one answer before it stops and reports.

    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.

    When you mention something discussed before, the assistant may look it up. Inside a project it searches only that project; incognito conversations are never searched.

    Saved notes are read at the start of each conversation, and new ones are saved when something is worth keeping. Off: nothing is read or saved, and the notes stay until you delete them.

    Health, religion, politics, ethnicity and sexuality are not saved unless this is on. Identity, passport, tax, bank and card numbers, criminal records and immigration status are never saved.

    Once a month the free Auto model reads your notes and suggests which to merge, remove or reconcile. It changes nothing: each suggestion waits below for you to accept or reject. Notes saved from your Google data are never sent.

    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.

    Delete all memory

    Saved notes

    System

    Synapsez version
    Keyboard shortcuts

    Every shortcut the app answers to.

    Browser use

    A browser on the assistant's cloud computer — it opens pages, clicks and fills them in, and you watch it work. Never your own browser.

    Computer use

    Its own computer in the cloud: it runs code there, makes files and installs what it needs. Never your own computer.

    Extensions

    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

    Detected tools

    Node.js: —

    Python: —

    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.

    The whole command line, exactly as you would type it in a terminal. This runs a program on your computer, so only paste something you trust.

    Need an API key?

    Every provider's keys — and the order they are tried in — are under Providers.

    Skills

    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.

    Connectors

    Plugins

    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.

    Recommended models

    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.

    Free allowances

    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.

    Feedback