GiState Privacy Policy

Effective Date: August 26, 2026 Last Updated: November 16, 2026 Operator: Dylan Ramirez, sole proprietor, doing business as GiState


1. Introduction

GiState ("we," "us," or "our") is committed to protecting your privacy. This Privacy Policy explains how we collect, use, store, and share information when you use the GiState API and platform (the "Service").

This policy complies with applicable 2026 privacy law including the Florida Digital Bill of Rights, GDPR, UK GDPR, CCPA/CPRA, California's DELETE Act, and Article 50 of the EU AI Act.


2. Information We Collect

2.1 Account Information

  • API key metadata (label, access mode, account balance in US dollars, safety fuse threshold)
  • Email address (if provided at registration)
  • Account creation and last-used timestamps
  • Your plan: whether you subscribe to Pro, its billing period, renewal date and status, and the Stripe identifiers for the subscription. Card details are held by Stripe, never by GiState
  • If you turn on two-factor sign-in: the secret your authenticator app uses, encrypted, and your backup codes, kept only in a form that cannot be read back
  • The channel you signed up from, when the link you followed was tagged or another site sent you (for example hn, or news.ycombinator.com). We use it only to count sign-ups by channel

2.2 Usage Data

  • Operations performed (refine, merge, rehydrate, checkout, branch)
  • Token volumes per operation
  • Amount charged per operation, in US dollars
  • Model tier used (frontier/standard/budget — not exact model name in logs)
  • Timestamps of all API calls
  • Safety Fuse trigger and reset events

2.3 Session Data (Gists)

  • Compressed AI session states submitted to the Service
  • Causal graph nodes and edges
  • Truth registry entries
  • Version control metadata (branch, depth, commit messages)
  • When you go back to an earlier checkpoint: which checkpoint, from which state, whether the goal was done or a dead end, the note you write, and how much came along. The state it writes holds what you chose to carry. Nothing is sent to an AI provider to do it

2.4 Technical Data

  • IP addresses (rate limiting and fraud prevention only, not retained beyond 30 days)
  • Request headers and API version
  • Error logs (retained 12 months)

2.5 Session Files

  • Named files that model replies write in a session (for example src/app.js), every saved version of each, which model wrote each version, and edits applied to them

  • Files you save, restore or delete yourself in the Files panel

  • Files of any kind you upload into a session (with the + under the message box, by dropping them on the conversation or by pasting them), under their file names, kept as session files like the rest: a text file as text; a document (Word, Excel, PowerPoint, OpenDocument, EPUB, RTF or HTML) as it is, with its text read out of it into a file beside it; an archive (zip, tar or gzip) unpacked into the files and folders in it; anything else as it is. The text of an uploaded file, and the name, kind and size of a file kept as it is (with the start of a long text file), are sent to the AI provider of the model that answers the message it was added for, and of later messages, as the session's other files are. Documents are read and archives unpacked on our servers; nothing is sent to an AI provider to do it

  • Images, PDFs, audio and video you add the same way are attachments to the message, kept as section 7 says for attachments. An attachment is sent to the AI provider of the model that answers the message it is attached to, which is always a model that reads that kind: PDFs go to Claude, Gemini and GPT models that see images; audio to Gemini models and OpenAI's audio models; video to Gemini models

  • A transcript of a recording you attach, when you ask for one with Transcript on the recording: the recording is sent to the AI provider of the model that writes it (one that listens to recordings), and the transcript is kept as a session file under transcripts/, which every model in the session reads

  • Files a program writes when it runs in the session's sandbox (section 2.7), kept as session files with the model that asked for the run

  • Word, Excel, PowerPoint and PDF files: one a model writes is made on our servers from the text it wrote, and both are kept as session files; one you download from a reply, a brief or a file is made on our servers from that text and not kept. Nothing is sent to an AI provider to make either

  • A brief you ask for with Make a brief in Pages: the text of the pages it reads (the session's answers as your browser shows them, with what was asked, their sources and their file names) is sent to the AI provider of the model that writes it, as a message is, and the brief is kept as a session file

Session files are stored on our servers with the session so they open on any device you sign in on, and so the next model in the session can read the current version. They are shown only to your account.

Pages are made in your browser from the conversation it already holds, and nothing is sent to do it. Which view each session opens in, and the newest brief made in it, are kept in that browser's storage.

2.6 Web Search and Page Reading

  • When a message uses web search, the search queries derived from it and the addresses of the pages read
  • If you turn on "Use the model's own search", the model's provider (Anthropic, OpenAI or Google) runs the searches itself as part of answering; we receive the queries it ran and the sources it cited, and keep them as above
  • The titles, addresses and dates of the sources a reply cites, kept with the checkpoint that saves the conversation
  • Search results and page text, kept briefly in a cache so a repeated search is not paid for twice (search results up to 6 hours, page text up to 1 hour)
  • If you turn on Research, a message gets more searches (up to eight, in two rounds) and more pages read, kept as above; for the second round, the small model that writes the searches also sees the titles, addresses and excerpts of what the first round found

2.7 Code Runs

  • When you run code from a session, the code and the session's other files (files kept as they are included) are sent to the code sandbox for that run
  • With the Sandbox switch on (under the message box, off until you turn it on), a model's reply, and each task of a Work job you run with it on, can ask to run code or one of the session's files in the session's sandbox. Each run sends what it runs and the session's files to the code sandbox, as above, and its output is sent to the AI provider of the model that asked, so the model can go on from it. Every model in the session works on the same files
  • Files a program writes when it runs are saved as session files (section 2.5)
  • A record of each run: the language, how it ended, its time, memory and charge, the model that asked for it and how many files it wrote. GiState does not keep the code or the output of a run you start yourself after the run. The runs a reply asked for, with their output, are part of the conversation (kept in your browser, and synced if history sync is on, section 2.8); the runs a Work task asked for are kept with the job

2.8 Chat History Sync

  • The conversation of each session (the messages, the models that wrote them, the runs a reply asked for with their output, and the checkpoints they stand on) and a short name for the device that saved it
  • On by default, and you can turn it off; see Section 3.6

2.9 Routing Signals

  • Which model answered a message, the kind of task it was classified as (coding, research, writing, extraction or reasoning), and whether the reply was retried, edited, continued or failed. The text of the message is classified and discarded; it is not stored.
  • Used only to improve which model Auto chooses, measured across accounts

2.10 Support Requests

  • What you send us through gistate.com/support, Help in the console or the API: your email address, an optional name, the topic, subject and messages, the page you came from (its address, without anything after a ?), and our answers
  • Whether you were signed in when you sent it, and if so which account it came from
  • Used only to answer you and to keep a record of what was asked and answered
  • A report of a bad handoff from the console also records which of five problems you picked and which model wrote the reply. Only if you tick "Attach" does it also keep a copy of the saved state, what the model was sent and the reply. Only support staff can open them, they are used only to find and fix the problem, and they are deleted when you remove them from the request, 90 days after the request closes, or with the request. Only if you also tick the second box may we make an anonymized test case from them, with names, numbers and anything identifying changed

2.11 Provider Keys (Optional)

If you add your own Anthropic, OpenAI or Google API key:

  • The key itself, encrypted and bound to your account
  • Its last four characters, to tell keys apart
  • Whether calls use it, and whether GiState's keys may be used at all
  • What the provider last said about it (accepted, refused, out of credit) and when it was last checked and used

2.12 Browser Extension (Optional)

If you install the GiState browser extension and add an API key to it:

  • It runs only on chatgpt.com, chat.openai.com, claude.ai and gemini.google.com, and reads a conversation there only when you press Save in it
  • What it saves is sent to GiState with your key and kept like any other saved session (Section 2.3); nothing else on those pages, or on any other site, is read or sent
  • Your API key is kept in that browser's extension storage, not by us; when you continue a saved state, the extension puts its continuation prompt into the site's message box, and nothing is sent to that site until you send it

2.13 Organizations and Teams (Optional)

If you make or join an organization (your company's workspace on GiState), or a team in one:

  • The organization's name, its members and their roles, its teams, and invitations by email until they are accepted, withdrawn or expire (7 days; we keep only a hash of each invitation link)
  • Your name and email address are shown to the organization's other members
  • You get a seat in the organization: a workspace it owns. Sessions you save while working in your seat are the organization's, not yours, and it pays for what you use there from its balance. When you leave or are removed, your seat and every session in it stay with the organization, and its owners and admins can hand them to someone else. Your own account, with its own sessions, stays yours and separate
  • A session shared with a team can be read by its members, with its saved states, sections and decisions, and continued in their own seats at the organization
  • A state shared from your seat with the organization's company gallery can be opened by its members only, with your name as the person who published it (Section 2.14). Continuing one puts the copy in their own seat at the organization. If the organization asks for it, an owner or admin approves a listing before anyone else sees it; owners and admins can feature, verify and take down listings and put them in collections, and you and the maintainers you or they add keep its tags and can mark it stale
  • The organization can set scan rules: words, phrases and hosts it does not want shared (client names, codenames, internal hosts). What you share from its seats is checked against them, and a match can stop the share or warn you. You see the rule's label, never what it matched; the audit log never has the words of a rule
  • A session you continued from a team's can be sent back to it as a proposal. Everyone on the team can read the proposal: your note, the decisions it changes and adds, and the saved state you sent. Whoever works in the shared session, or a team owner, accepts or declines it; accepted, your state, or the decisions they choose from it, become part of that session, in the organization's seat it lives in. We keep who proposed, who decided, when and the note, until the organization is deleted; your name is shown from your account, so it goes when your account does
  • Members of a team see every decision in its shared sessions in one list, with whose session it is or who proposed it, whether the person or the assistant said it, and the words quoted from the conversation; and can search the saved states of its shared sessions. That search never reads the conversations kept in a session's archive. Opening that list and searching are in the audit log, without what was searched
  • Each team has a room: a thread its members post in, and the team's goals, decisions and dates beside it, each with why, an owner and a day where it has one, and who added, approved, rejected and changed it. Everyone on the team reads it; your name is shown on what you post and change. A message you post is deleted when you or a team owner or admin delete it; the lines the room writes for each change, and the items themselves, stay as the room's history until the team is deleted. When a proposal to one of the team's sessions is accepted, the decisions it brought in are added to the room, with who accepted it; when one is declined, the room says so, with the note. When a Work job you run finishes in a session the team shares (or one continued from it), and you are a member of the team or above, the room gets a line from you with the job's goal, whether it finished or failed, how many of its tasks did, what it cost, the models it used and the start of its answer or its error; like the room's other lines, it stays as the room's history until the team is deleted. We keep when you last read the room, to count what is new
  • When a member works in a session shared with a team, or continued from one, the team room's current goals, decisions and dates, with the names of their owners and of who approved them, are sent to the AI provider of the model they use with each message, as part of what the session sends (Section 5)
  • The organization's owners and admins see an audit log of what is done in it: who did what and when, and from which IP address: invitations, people joining, role changes, removals, seats handed on, teams and sessions shared, states shared from its seats (and with whom), team states opened, proposals sent, opened, accepted, declined and withdrawn, items added to a team room and changed (never their words), the decision list opened and searches across a team, company gallery listings published, approved, turned down, featured, verified, changed and continued, a newer state held back from one, collections and scan rules changed, states exported, and changes to the balance. It never holds what a session says. They can export it and have it sent to their own systems
  • The organization sets rules for its seats: who may share out of them and for how long, how long its states are kept (older ones are deleted automatically), and legal hold, during which nothing in its seats is deleted, even when you ask, until its owners lift it. Its owners and admins see what each seat was charged and on which sessions, and can cap it
  • If the organization pays by invoice, the billing address and company name its admins give go to Stripe, which sends the invoice
  • If you delete your GiState account, nothing of the organization's goes with it: your seat goes to one of its owners, and your name and email address are taken out of its audit log
  • An organization can verify its email domains and sign its people in through its own identity provider (single sign-on). When you sign in that way, the identity provider tells us your email address and name; we make your GiState account if you have none, mark the session as made through that sign-in, and keep nothing else it sends. If your organization requires single sign-on, an address at its domains signs in only through it. If your account has two-factor on, we still ask for your code
  • An organization's identity provider can add, update and remove its people on GiState (SCIM). For each person it provisions we keep the username, email address, name and directory ID it sends, and whether it says the person is active, and the groups it puts them in become teams. When it deactivates or removes you, you leave the organization: your seat goes to one of its owners and the sessions its single sign-on made for you end. Your own account stays yours

2.14 Sharing and the Gallery (Optional)

If you share a saved state, or publish one to a gallery:

  • Whoever it is shared with can open it: the people you invite (signed in at those addresses), whoever has the link (and its access code, if it asks for one), approved GiState members through the Gallery, or the members of your organization through its company gallery (Section 2.13). They see the whole state, its title, your note and its tags. In the Gallery your name is not shown; in a company gallery it is
  • Before a share is made, the state is checked for what reads as private (keys, tokens, card numbers, email addresses and phone numbers) and, in an organization's seat, for its scan rules. You are told what kind of thing was found and where, never the text, and nothing of the check is kept
  • If you turn on "Keep it up to date", the listing follows its session: each newer state you save there is shown in its place once it passes the same checks, and we keep which states it showed, with what kind of thing the checks warned about in each and where. A newer state that does not pass is not shown; we keep what kind of thing held it back and where, for you and whoever looks after the listing to see
  • When someone continues a listed state, we record which account copied it, from which state and when, so they can be told what changed in its decisions when the listing moves on, and take what they choose into their copy. Whoever published the listing sees how many times it was continued, by how many people, and how many bad handoffs were reported on those copies, never who

2.15 Memory (Optional)

Memory is off unless you turn it on (Settings → Memory). When it is on:

  • We keep short notes about you, each one sentence: who you are and your work, how you like answers, preferences and constraints, ongoing projects, the tools you use. Each note records where it came from: learned in which session and at which checkpoint, and by which model; added by you; or imported by you
  • Your notes are sent with every message and every Work task to the AI provider of the model that answers, whichever model that is, so it knows them too
  • With "Learn from my sessions" on (it is, once memory is on, until you turn it off), after each checkpoint the conversation turns that checkpoint saved, and the notes already kept, are sent to the AI provider of a low-cost model, which suggests notes to add, change or remove; we apply them. It is billed to you like a short reply. It is told not to keep secrets, details that only matter to one task, what a model or a document claimed rather than you, or sensitive details (health, beliefs and the like) unless you asked for them to be remembered. A model never rewrites or removes a note you added or imported
  • What a session learned is shown in that session after the checkpoint, and every note is listed in Settings, where you can edit or delete any of them, add your own, import a list (pasted; nothing is sent to a model to import it), export them all, or delete them all at once. Turning memory off stops sending and learning; your notes stay until you delete them
  • In an organization seat, the seat keeps its own memory, separate from any other account

2.16 Assistants (Optional)

If you make an assistant (a model, instructions, files and suggested tools set up for a kind of work):

  • We keep what you give it: its name, what it is for, its instructions, the model you chose, its conversation starters, the tools it suggests, and its files (with the text read out of documents and the files unpacked from archives, as for a session's uploads), with who made it and when
  • It is yours alone unless you share it from an organization seat with a team you are in or with everyone in the organization. Then they see all of it, its instructions and files included, with your name as its maker, from their seats at that organization, and can start sessions from it; only you can change it, and you or the organization's owners and admins can delete it
  • Starting a session from an assistant copies its files into the new session, as the session's own: they stay there, under that session's rules (Section 7), even if the assistant is later changed or deleted. The assistant's instructions are read again with every message and every Work task in that session and sent to the AI provider of the model that answers, while the assistant still exists and is still shared with whoever works in the session
  • A session started from someone else's assistant is yours (or your seat's), and billed like any other

2.17 Connectors (Optional)

If you connect one of your own services (GitHub, Linear, Notion, Jira, Stripe and others) through its MCP server:

  • We keep its address, the token you give us or the tokens your sign-in at the service gives us (sealed, never shown again), GiState's registration at the service, the list of its tools, and your choices for them
  • Signing in happens at the service: we see what it sends back to finish the sign-in, and keep the tokens
  • Nothing is sent to a model until you tick the connector in a conversation's tools. Then its tools' names and descriptions go with each message to the AI provider of the model that answers (Section 5)
  • When a model calls a tool, we send the call (the tool and the arguments the model wrote, or that you edited) to the service, with your token, when you run it, or at once for a tool you let run without asking. We keep each call: the tool, its arguments, whether it ran, and what the service answered, cut to a length, shown in the session
  • What the service answered goes to the AI provider of the model in the next message, so it can go on. The services you connect are yours to choose, and their own terms and privacy policies apply to what they receive
  • Removing a connector deletes its token or sign-in here; to revoke GiState's access at the service as well, do it in the service's own settings

2.18 Scheduled Work and the Daily Brief (Optional)

If you schedule work:

  • We keep what you set: its name, the question or the job's goal, the session it is about, the model or mode, the most a run may cost, when it runs and in which time zone, and whether to email each result
  • When it runs, we send what a chat turn or a Work task would send (the question, the session's newest saved state, its assistant, your standing instructions and memory) to the AI provider of the model that answers (Section 5), and charge your balance as for any reply
  • We keep each schedule's newest 50 runs: when it ran, what it cost, the model, the answer (cut to a length) or the Work job, and why it failed or was skipped. An answer is also kept in the session's files, under that session's rules (Section 7)
  • If you asked, each result is emailed to your sign-in address through our email provider

If you turn on the daily brief:

  • Once a day we put together, from your own records and without sending anything to an AI provider, a summary of your sessions' checkpoints, Work jobs, scheduled runs, what waits for your approval, what memory learned, what you spent and your balance
  • We keep each brief so Home can show it, and email it to your sign-in address through our email provider unless you turn email off

2.19 Sidebar Sections (Optional)

If you make sections in your sidebar, we keep their names, their order, whether each is collapsed, and which section each session is in. Deleting a section only moves its sessions back to Recent.

2.20 Files From Google Drive (Optional)

If you choose Add from Google Drive:

  • You sign in with Google in Google's own window and pick files in Google's own picker. Google handles that under its own privacy policy
  • We ask Google only for access to the files you pick (Google's "drive.file" permission), never to the rest of your Drive
  • The access Google gives is kept in your browser tab's memory for up to an hour and is never sent to our servers. Sign out of Google Drive in the menu forgets it and asks Google to revoke it
  • The files you pick are downloaded from Google into your browser and then added to your message or your session's files exactly as if you had added them from your computer, under the same rules (Sections 2.5 and 7)

3. How We Use Your Information

Purpose Data Used Legal Basis
Providing the Service Session data, usage data Contract
Billing Usage data, account info Contract
Safety Fuse enforcement Session spend rate, session data Contract / Legitimate interest
Security and fraud prevention IP, usage patterns Legitimate interest
Legal compliance All data as required Legal obligation

3.5 Session Archive (Opt-In)

By default GiState stores only compressed state — the structured gist produced from your session. The original session text is processed to create that gist and is then discarded.

If you enable the Session Archive, GiState additionally retains the raw session text you submitted, so exact details that compression does not preserve remain searchable. This is a meaningful change to what we hold about you:

  • Off by default. Archiving is enabled per API key and is never turned on for you.
  • What is stored: the verbatim session content you submit to /refine, linked to the gist created from it.
  • Retention: you choose a retention window (1–3650 days, default 90). Archived text is deleted automatically once it expires.
  • Deletion: you may delete archived text at any time (DELETE /sessions/:g_id/archive) while keeping the compressed gist. Deleting a gist also deletes its archived text.
  • Access: archived text is subject to the same account-level access controls and row-level security as your gists.
  • Search: retrieval uses database text indexing only. Archived content is not sent to any third-party AI provider as part of a search.

Because the archive retains raw content, consider whether the sessions you submit contain personal data, credentials, or client-confidential material before enabling it.

3.6 Chat History Sync

GiState keeps each session's conversation on its servers, alongside the compressed state of each checkpoint (Section 3.5), so every device you sign in on opens a session with its messages:

  • On by default. You can turn it off at any time in Settings → Chat history on every device. With it off, the text of a conversation stays only in the browser it was written in.
  • What is stored: one copy per session of the conversation as the console holds it (up to 400 turns and 2 MB), with the device that last saved it.
  • Deletion: turning history sync off deletes every saved conversation at once. Deleting a session deletes its copy.
  • Access: saved conversations are shown only to your account and are not sent to any AI provider except as part of a message you send.

3.7 Session Files, Web Search and Code Runs

  • Session files are kept until you delete the file, the session or your account. Each file keeps its first version and up to its 100 most recent versions; older versions are removed as new ones are saved.
  • Web search sends search queries to the search provider configured for the Service (Brave Search, Tavily or Exa) and fetches pages directly from the sites they are on. Text from web pages is given to the model as untrusted reference material, labelled with its source.
  • The model's own search (off unless you turn it on) lets the provider of the model you are using search the web while it answers: Anthropic's web search, OpenAI's web search, or Grounding with Google Search. The provider, and the search services it uses, receive the queries the model writes from your message, under that provider's own terms. Answers grounded in Google Search show Google's search suggestions, as Google requires.
  • Code runs happen in an isolated sandbox with no network access and limited time and memory. Run records are deleted after 90 days. Models run code there only with the Sandbox switch on, and the files a run writes are kept as session files, like the session's other files.
  • Routing signals are deleted after 90 days.

4. No Training on Your Content

GiState does not use your content (your messages, conversations, saved states, files or images) to train or fine-tune AI models, and keeps no copies of it, anonymized or otherwise, for that purpose. The routing signals in Section 2.9 hold no message text. How the AI providers that GiState calls on your behalf treat what they receive is described in Section 5.


5. Third-Party AI Providers

GiState calls third-party AI model providers on your behalf. By default it uses its own credentials with those providers.

If you add your own provider key (Settings → Provider keys) and choose to use it, calls to that provider are made with your key, under your account with that provider. Those calls are then also governed by your agreement with the provider and by its data practices for your account, including any logging or retention you have turned on there, and your provider account handles the requests GiState sends on your behalf, GiState's own instructions to the model included. Your key is used only for calls made for your account.

When you run an operation that requires inference, the relevant portion of your session content is transmitted to the provider selected for that request. In relation to those providers, GiState:

  • Uses enterprise/API-tier access, which typically excludes customer data from the provider's own model training programs
  • Sends only the content required to perform the requested operation
  • Logs metadata about the call — model, token counts, cost, latency — for billing and reliability purposes
  • Does not control, and cannot guarantee, any provider's data practices

Some features send the same content to more than one provider, when you turn them on (the Sidecar, Fusion and Challenge switches under the message box) or let them run automatically (Settings → Sidecar and Fusion):

  • Sidecar sends your message and the reply it checks to a model from another provider, which writes a short check of it
  • Fusion sends your message, with what the session sends with a message, to several models, usually from different providers, and then their answers to the model that writes one answer from them
  • Challenge sends your message, the reply and recent turns to two or three models from other providers

Whether Sidecar and Fusion run only when turned on or also automatically, and the most what runs automatically may cost on one message, are kept with your account; whether their switches are on is kept in your browser. In an organization's seat, its owners and admins set the most its seats may use. Deciding what each does for a message, and whether a message calls for it, reads the message and sends nothing to any provider.

The providers we use as sub-processors are:

Provider Purpose
Anthropic AI model inference
OpenAI AI model inference
Google AI model inference
Brave Search, Tavily or Exa (whichever is configured) Web search queries, when a message uses web search
Anthropic, OpenAI or Google (the model's provider) Web search queries, when you turn on the model's own search
Judge0 (self-hosted or through its hosting provider) Running code you choose to run
Render Application hosting and database
Cloudflare DNS, CDN, marketing site hosting, and storage of the images, PDFs, audio and video you attach
Stripe Payment processing
Resend Transactional email
Sentry Error monitoring

We will give notice before adding a new sub-processor that processes customer content.

We recommend you review the terms of any provider whose models you choose to route requests to.


6. Data Storage and Security

  • All Gist data stored in encrypted PostgreSQL with AES-256 at rest
  • Row-Level Security (RLS) at the database level — your data is inaccessible to other users even in the event of an application-layer bug
  • API keys stored as SHA-256 hashes — GiState cannot recover your raw key
  • Provider keys you add encrypted with AES-256-GCM, bound to your account, decrypted only to make a call for your account, and never shown in full again
  • All data transmitted over TLS 1.3 minimum
  • Data stored in the United States (Florida/US-East region)
  • Independent security audit: not yet performed. We will update this policy when one is completed. We do not currently claim SOC 2, ISO 27001, or equivalent certification.

7. Data Retention

Saved states and checkpoints are retained until you delete them or your account is deleted, subject to applicable legal requirements. Optional raw-transcript archives (the Session Archive, Section 3.5) and certain operational records have separate retention periods, described below.

Data Type Retention Period
Saved states and checkpoints Until you delete them, their session or your account. Inactivity never deletes them
A record of going back to an earlier checkpoint, and its note (Section 2.3) Until its session is deleted
Images, PDFs, audio and video you attach Until every saved state using them is deleted, then removed within two days. An attachment never saved into a state is kept until you delete your account
Archived session text (opt-in, Section 3.5) The retention period you choose (default 90 days), or until you delete it
Session files, including uploads, the text read out of documents, transcripts and files written by runs Until you delete the file, the session or your account
Synced conversations Until you turn history sync off or delete the session
Cached search results and page text Up to 6 hours
Code run records and routing signals 90 days
Support requests 2 years after the request is closed, or until you delete your account
What a handoff report attached (only if you ticked Attach) 90 days after the request is closed, or until you remove it from the request
Provider keys you add Until you remove the key or delete your account
Memory notes and the memory setting (Section 2.15) Until you delete them (one by one, or all at once) or your account. Deleting them erases them at once
Schedules and their runs, with the answers (Section 2.18) A schedule until you delete it or your account; its newest 50 runs. Answers kept in a session's files follow the session
Sidebar sections (Section 2.19) Until you delete the section or your account
Daily briefs and their settings (Section 2.18) Until you delete your account
Connectors: their tokens and sign-ins, tools and settings (Section 2.17) Until you remove the connector or delete your account; a sign-in under way, ten minutes
Connector calls: the arguments and what the service answered (Section 2.17) Until you delete your account
Assistants and their files (Section 2.16) Until it is deleted (by you, or, when shared in an organization, by its owners and admins), or the account it was made in is (one made in an organization seat stays with the seat). Copies of its files in sessions started from it stay with those sessions
Audit logs 12 months
An organization's audit log (Section 2.13) Until the organization is deleted. Your name and email address leave it when you delete your account
A team room: its items, the lines it writes for each change and for each Work job finished in the team's sessions, and when each person last read it (Section 2.13) Until the team is deleted. Your name leaves them when you delete your account
A message you post in a team room (Section 2.13) Until you, a team owner or admin delete it, or the team is deleted
A proposal to a team's session (Section 2.13) Until the organization, the team or either session is deleted. Who sent and decided it is kept as your account, so your name leaves it when you delete your account
Sessions in an organization seat (Section 2.13) Until the organization deletes them, its retention period passes, or the organization is deleted; never while it is on legal hold
What an organization's identity provider sent about you through SCIM (Section 2.13) Until the identity provider deletes you, your account is deleted, or the organization is deleted
A single sign-on in progress (Section 2.13) One day after it expires (it is good for ten minutes)
An organization's scan rules (Section 2.13) Until its owners or admins remove them, or the organization is deleted
A share or Gallery listing, its tags and marks, and the states a listing showed (Section 2.14) Until it is deleted: with the saved state it shows, or with the account that made it. Revoking it stops it at once
Who continued a listing, and from which state (Section 2.14) Until the listing or the copy's session is deleted
Usage ledger 24 months
IP address logs 30 days
Account data Until you delete your account, or 30 days after GiState terminates it

When you delete something listed above, it is removed from GiState's systems straight away (an attachment within two days, once no saved state uses it). Copies can remain in the database host's backups for up to 30 days before they are overwritten. Deleting your account keeps only the record of what you were charged and paid, including what Stripe sent us about each payment, for tax and accounting.


8. Your Rights

Depending on your location, you may have the following rights. Submit requests to support@gistate.com. We respond within 30 days (15 days for California residents under CPRA).

Right Description
Access Request a copy of data we hold about you
Correction Request correction of inaccurate data
Deletion Request deletion of your account and Gist data
Portability Request a copy of your Gist data by contacting us
Restriction Request restriction of processing
Human review Request human review of any automated decision affecting your account

Florida Digital Bill of Rights: Florida residents have the right to opt out of the sale of personal data (we do not sell data), correct inaccurate data, and obtain a copy of their data.

GDPR/UK GDPR: EU/UK residents have additional rights including the right to lodge a complaint with your local supervisory authority.


9. Data Sharing

We do not sell personal data. We share data only in the following limited circumstances:

  • Infrastructure sub-processors: Under strict data processing agreements (DPAs)
  • Legal requirements: Court orders, government authority, or applicable law
  • Business transfer: Merger or acquisition with 30 days notice to affected users
  • Organizations you join: their owners and admins see what is done in their workspace in its audit log, and the sessions in their seats (Section 2.13)
  • With your explicit consent: For any other purpose

The current sub-processor list is in Section 5 above.


10. Children's Privacy

The Service is not directed to individuals under 13. We do not knowingly collect information from children. Contact support@gistate.com if you believe we have.


11. International Transfers

Data is processed in the United States. For EU/UK users, transfers rely on Standard Contractual Clauses (SCCs). For other international users, by using the Service you consent to transfer to the United States.


12. Changes to This Policy

We will provide 30 days notice of material changes via email. Continued use after the notice period constitutes acceptance.


13. Contact

GiState — Privacy Dylan Ramirez, sole proprietor, doing business as GiState 142 West Thatch Palm Circle, Jupiter, FL 33458 support@gistate.com