About and trust
An agent can do everything except move money.
That single sentence is the product. Everything else — the API shape, the free tier, the wallet, the colour of a button — follows from it.
Why the invariant, and not a policy
Plenty of providers say an agent "shouldn't" spend without approval. Cloudflare's own registrar documentation puts the responsibility on the developer to design a flow that will not buy domains unsupervised. That is a reasonable thing to write and a bad thing to depend on, because it means the guarantee lives in your prompt rather than in our server.
Here it lives in our server. Every debit in the system goes through one function. There is no second path, no "just this once" flag, and no card on file to fall back to. An order either has a human authorisation artifact behind it — a paid pay-link, or a live wallet mandate with headroom — or it does not execute.
the only human step
An email with line items and a total. Approve, or close the tab. Nothing is charged and nothing is provisioned if you close the tab, and an unapproved order expires by itself.
The wallet is a mandate, not a stored card
When you approve the first purchase you can add credit and set a per-purchase cap. That cap is enforced server-side. Your agent can spend underneath it without emailing you; anything above it comes back to you as another pay-link. You can empty the wallet, lower the cap, or revoke the token at any time, and we warn you at 30 days of remaining runway rather than at zero.
This is deliberately not "a card on file with good intentions". There is no card. There is a balance you put there and a ceiling you chose, and the worst case for a compromised or confused agent is bounded by a number you picked.
What we are honest about
- Cloudflare is cheaper than us on domains. They resell .com at their own wholesale price, which is below what ours charges us, so we cannot match them without losing money on every domain — and we will not do that. Ours is $12.49: $11.50 registrar, $0.66 card fee, $0.33 us. The full breakdown is on the pricing page and in pricing.json.
- Our first plan ladder was underpriced and we cut the allowance instead of raising the price. The draft promised 100 GB of bandwidth on the $7.00 plan, which loses money as soon as a customer uses it. The allowances are now counted at the origin and every paid plan covers its own worst case. We are recording this because the published prices are the second attempt, not the first, and the pricing page carried a blank rather than a number we would have had to raise.
- Free plans throttle, they do not bill. Exceeding a free quota slows your app down. It does not produce an invoice.
- Cold starts are real. Scale-to-zero means a couple of seconds on the first request after idle. We sell a warm plan rather than pretending the cold one is fast.
- Renewals cost what registration cost. No promotional first year.
- We are not the cheapest at anything, and we do not claim to be. On domains Cloudflare is cheaper than us and always will be — their wholesale price is lower and they add nothing to it. If price is the only axis that matters to you, several of the providers on our comparison pages are fine choices and we say so on those pages.
What is actually built, as of 17 September 2026
Slipdock is live. Domains register against the production registrar; cards are charged. There are still no paying customers. No number on this site was measured on production traffic, no testimonial exists, and no logo belongs to anyone but us. Here is the real state.
works end to end in production
- Workspace creation and instant scoped token
- Free private git repository
git push→ buildpack build → running app serving HTTP 200- DNS zone with A, TXT, MX and CNAME records resolving
- Quota enforcement and the dunning ladder
- Domain search and a completed registration against the registrar sandbox
- Wallet, ledger, idempotency, the
next_actionerror contract
Not done, and visible if you look
- Transactional mail is not domain-branded yet. Pay-links
and receipts arrive from
slipdock@nativefoundation.ai(our parent brand's verified domain) whileslipdock.ai's own DKIM/SPF setup is pending. Deliverability is fine today; the From address is the obvious tell. - Hosted Postgres is on the roadmap, not live. The T2 managed-database tier is designed and priced but not shipping; agents get free git, static hosting and app deployment today, not a database.
- VMs are deferred. No drive-as-compute yet — apps run on the shared Cloud Run platform, nothing more privileged.
- abuse@slipdock.ai does not exist yet. Suspect abuse is
reportable via
/v1/abuse/reports(agents) and every registration passes an LLM abuse screen, but a human-addressable mailbox is not on yet and it should be. - Launch-week unknowns are unknowable. No paying customers yet means no scale-tested numbers — our published caps are engineering limits, not battle scars.
The thing that could actually kill this
Not competition — abuse. An instant token, a free host and a cheap domain is exactly what a phishing operation wants, and the real failure mode is our upstream registrar terminating us because of what somebody else did with our API.
The human payment anchor is the primary defence: every workspace that ever buys anything has a verified human payer attached to it. On top of that, nothing gets registered before funds clear, velocity limits apply, and published hostnames are screened before they become publicly reachable. If you find something abusive hosted by us, we want to hear about it, and the reporting route will be published before we take a single live sale.
Who runs this
Slipdock is built by NativeFoundation. It is a commercial product and it intends to make money — on hosting, databases and the wallet, not on surprising you at renewal.
Leaving must always be easy
This is the second of our written principles and it ranks immediately after the money invariant, because the two are the same argument. You can export your repo, your database, your DNS zone, your billing history — and anything else we ever hold on your behalf — at any time, in a standard format, and walk away with it. No export fee, no "contact sales", no degraded export, no hostage data.
| What you have here | How you take it with you | Costs? |
|---|---|---|
| Your code | The repo is a plain git remote. git clone and you have everything, including history. | nothing |
| Your database | pg_dump, on demand, including while a workspace is suspended for non-payment. | nothing |
| Your DNS | Exports as a standard zone file. Point the domain at any nameservers you like — we do not require you to use ours. | nothing |
| Your domain | Transfer auth code on request. No fee, and no hold beyond the one ICANN mandates. | nothing |
| Your billing history | The full ledger, exportable. Every debit, with its reason. | nothing |
| Your users, if we ever hold them | With their password hashes, in a standard format. We are not in the authentication business today; if that changes, this row does not. | nothing |
Why write this down rather than just meaning it. We are an aggregator, and aggregators accrue switching cost naturally: once someone's domain, code, database and DNS all live in one place, leaving means re-integrating four things. That gravity is earned and legitimate, and we are happy to have it. Friction is different — it is a decision to make the exit worse, and every company that makes it tells itself it is protecting revenue. The moment we do that we stop being the trustworthy option, which is the only reason an agent ecosystem would pick us over an incumbent with more money.
There is also a cold-blooded version of the same point. Our buyer is an agent: it has no sunk-cost bias, no embarrassment about cancelling, and perfect recall of what we promised. It will re-run the comparison at every renewal. Lock-in does not work on it, and attempting lock-in becomes a machine-readable fact about us that outlives the customer we tried it on.
The principles this is built on
Short list, and load-bearing: a change that violates one of these is a change to what the product is, not an implementation detail.
- An agent can do everything except move money. One function in the codebase moves money and nothing else may debit the ledger.
- Leaving must always be easy. Standard-format export of everything, any time, no fee.
- Errors say what to do next. Every non-2xx carries a
next_action; a 402 is a normal step, not a failure. - Nothing is silent. No silent deletion, billing, truncation or downgrade. Given a choice between quiet convenience and loud honesty, choose loud.
- Caps are enforced, not advisory. Free limits throttle rather than bill; wallet caps are checked before money moves.
- Deterministic surfaces. Idempotency keys on every mutation, nothing requires JavaScript, the same nouns in REST, MCP and HTML.
- Prices do not bait. Renewal equals registration, always, and every domain price is published broken down into registrar cost, card fee and our margin.