Security and trust
What's in place, what's next, and what we don't claim
You'll be asked these questions by IT and your auditors, so here are straight answers. Kipso is an early preview: a working demo with fictional data. It isn't in use on real projects yet.
How the product is designed today
These are built into the working demo.
- People approve; the agent prepares. The agent drafts and checks. Approvals, certifications and publishing are done by named people, and certification is blocked while blocking problems remain.
- Approvals follow your policy. Only the person a change order is waiting on can act on it, and a person who approved an earlier step can't approve a later one.
- A record of who decided. Approvals, certifications and agent actions are recorded with the person and time in an append-only, hash-chained audit log. Auditors can export it and check that no past entry was changed or removed.
- People sign in as themselves. Sign-in is by single-use email link, and permissions follow each person's role. The public practice sandbox, and a workspace's setup key until an admin turns it off, use names as given instead, and the log labels them that way.
- The public page shows only what staff published. The bond page is server-rendered HTML with no scripts, and shows only published text and headline figures.
- AI keys stay on the server. The app calls the AI model from our server, so keys never reach the browser. The demo is hosted on Cloudflare.
- Outside AI assistants are opt-in, per person. See AI and your data below.
- No guessing on money. Pay-app lines that can't be matched with confidence go to a person instead of being filled in.
AI and your data
Your IT team will ask what AI touches your data and whether outside AI tools can reach it. Here's exactly how it works in the demo today.
The AI inside Kipso
- One AI provider: Anthropic. Reading pay-app PDFs and setup files, drafting change orders, updates and the board narrative, and the in-app assistant all use Anthropic's Claude models, called from our server.
- Only what the task needs is sent. For example, the pay-app PDF being checked, the setup files you upload, or the project figures a report is drafted from.
- We don't train models on your data.
- Agent actions are logged. Each action the in-app assistant takes is recorded with the tool it used, the model and the reason it gave.
Outside assistants (Claude, ChatGPT and others)
- Off until a person turns it on. An assistant can connect only after a signed-in person creates a personal access token for themselves. An admin can also switch off the workspace's setup key, the one other way to connect.
- Scoped to that person. The assistant acts as the token's owner, with their role: read-only people get read-only tokens, and approval rules and separation of duties apply as they do in the app.
- Sensitive actions are confirmed in Kipso. Approving, certifying, publishing or changing people and rules from an assistant only proposes the action. The person confirms it in the app.
- Revocable. Tokens expire within a year, can be revoked at any time, and are revoked automatically when someone is removed. We keep only a hash of each token.
- Every change is in the audit log. Changes made through an assistant are recorded with the person and time, in the same hash-chained log as the app.
- Your assistant's provider sees what it reads. Data your own assistant reads is processed by that assistant's provider under your account's terms, not ours.
These controls are built and tested in the demo. Production hosting for our first pilots is being set up, and we'll confirm the terms in writing before you share real data. We hold no security certifications yet.
Kipso Bridge and your files
Bridge handles the files you migrate, so it has its own controls. Private workspaces for real projects are built and being tested. They aren't open yet, and won't hold real data until a data agreement is signed.
Private workspaces for real projects
- Sign-in by email link. Links work once and expire after 15 minutes. Membership is checked on every request, and organizations are kept apart.
- The data agreement comes first. Everyone accepts it before any project or file is opened, and the acceptance is recorded.
- Files stored privately, encrypted at rest by our hosting provider, with no public access. Uploads are checked by file type and size.
- A hash-chained audit log of uploads, loads, undos and downloads. Any staff access is written to your log too, and never includes file contents.
- Deletion on your schedule. Source files are deleted a set number of days after a project is complete, or right away. Deleting everything leaves only a receipt, which we email to you.
Who processes your data
- Cloudflare: hosting and file storage.
- Anthropic: reading documents and suggesting matches. We don't train models on your data.
- Resend: sign-in links, invitations and receipts.
- Stripe: payments, once paid migrations are offered.
Not built yet
Two-factor sign-in, malware scanning of uploads, and a choice of where your data is stored. The public Bridge demo is for sample files only, so please don't upload confidential data there.
Roadmap
In the order we plan to build it. We'll update this page as each item ships.
| Item | What it means | Status |
|---|---|---|
| Separate workspaces per organization | One organization can never see another's data | Before first pilot |
| Sign-in and roles | Each person signs in as themselves, with permissions by role | In the demo |
| Approval rules each owner configures | Dollar tiers and policies like "only the director releases contingency" | Before first pilot |
| Sandbox per customer | Train and test outside production | Before first pilot |
| Audit trail screen and export | History of each record with the person and time, agent actions marked, exportable for auditors | In the demo |
| SOC 2 Type 1, then Type 2 | Independent audit of our security controls | Planned; work starts during pilots |
| Data retention settings | Keep records as long as your retention schedule requires | Planned |
What we don't claim
- We are not SOC 2 certified yet.
- We are not FedRAMP authorized and don't plan to pursue it soon; our focus is public agencies and private owners.
- We have no production customers yet. Screenshots on this site are from a demo with a fictional district.
Have a security questionnaire? Send it with your demo request, or email it to hello@getkipso.com. We'll answer it honestly, including "not yet" where that's the answer, and tell you when each gap is scheduled to close.
Common questions
Where is our data stored?
The demo is hosted on Cloudflare and uses only fictional data. Before you share any real data, we'll put in writing where pilot data is kept and on what terms, and each organization gets its own workspace before the first pilot.
Is our data used to train AI models?
We don't train models on your data. The app sends documents to our AI provider, Anthropic, to produce drafts and checks. When you use Kipso from your own AI assistant, that assistant's own data terms apply.
Can outside AI tools like ChatGPT or Claude reach our data?
Not unless a person turns it on. An outside assistant can connect only after a signed-in person creates a personal access token for themselves, and an admin can switch off the workspace's setup key, the one other way to connect. The assistant then acts only as that person, within their role, and sensitive actions such as approving or certifying are only proposed until the person confirms them in Kipso. Tokens expire within a year, can be revoked at any time, and are revoked automatically when someone is removed. Every change made through an assistant is recorded in the audit log.
Which AI provider processes our data, and what's sent?
Anthropic. Kipso's AI features send what each task needs, such as the pay-app PDF being checked, the setup files you upload or the figures a report is drafted from, to Anthropic's Claude models from our server. We don't train models on your data.
Is Kipso SOC 2 certified?
No, not yet. SOC 2 Type 1 and then Type 2 are planned, with the work starting during pilots, and we'll update the security page as each step is done.
Who can approve and publish?
Only the people your approval policy names, and certification is blocked while blocking problems remain. Approvals, certifications and agent actions are recorded with the person and time in an append-only, hash-chained audit log. Signed-in people are recorded as themselves; actions taken with a workspace's setup key or in the public practice sandbox are recorded under the name given and labelled as not signed in.
What happens to our data if we leave?
It's yours. A self-serve export isn't built yet, so before a pilot we'll agree in writing how you get a full copy of your records and how we delete your data after you leave.
Can we send our security questionnaire?
Yes. Send it with your request and we'll answer it honestly, including "not yet" where that's the answer, with when each gap is scheduled to close.
What does our IT team need to do?
Usually a review, not a project: security and data handling, and the accounting export, a CSV mapped to your system's import format, imported by your finance team. Sign-in is by email link today; single sign-on isn't built yet, so tell us what you use. For moving data in, Kipso Bridge works from exports your teams already produce.