Your systems. Clear limits. Every action on record.
AI that touches your phones, pricing and invoices has to be built with boundaries. Here's exactly how we scope access, keep your people in control, and what we will — and won't — claim.
Agents see the inboxes, systems and records we list together — nothing else.
02Scoped per system, per action
What it can do
Separate credentials for each connection, limited to the actions the job needs.
03Approval + audit trail
What needs a person
Quotes, pricing, money and anything unusual wait for your team, and every step is logged.
01 /
Where it runs and how data flows
Before we build, you get a plain-English map of every system we connect, what data moves between them and where it's processed.
Each build documents its source systems (phones, inbox, CRM, ERP, accounting), the external services involved and the data each connection uses.
Hosting region: US East (Virginia). Data at rest and in transit is encrypted by the hosting and database providers.
Connecting a system doesn't mean copying it. Agents read what a task needs, when it needs it, and we keep only what the work requires.
We say plainly that some data does leave your systems: AI model providers and connected platforms process the context they're given under their own terms. The data-flow map shows exactly which.
02 /
Access: least privilege by default
An agent that answers your phones doesn't need your bank login. Every connection gets only the permissions its job requires.
Separate, scoped credentials per system and per action, preferring read-only wherever the job allows.
Secrets are kept in a managed secret store, never in code, spreadsheets or chat.
Your team can see what each agent can access, and access can be revoked system by system without taking everything down.
dev1 staff access is limited to the people running your build, and is removed when it's no longer needed.
03 /
People approve what matters
Automation handles the routine. Anything consequential stops and waits for a named person on your team.
Approval points are agreed in writing during the build: quotes, pricing changes, refunds, credits, unusual customer requests.
Agents hand off to a person the moment a conversation goes outside what they've been set up to handle.
Approvals are fast — one tap from the dashboard, text or email — so control doesn't slow the work down.
04 /
Every action is logged
If an agent did something, you can see what, when, why and who approved it.
Every message sent, record changed and document produced is written to an activity log.
Approvals record the person and the time.
Logs are available to you on request and on offboarding. Retention: 12 months.
05 /
AI models and your data
We use commercial AI models through their business APIs, not consumer chat apps.
Model providers: Anthropic (Claude). We use business API terms under which customer data isn't used to train their models — confirm per provider in your data-flow map.
We send models the context a task needs, not your whole database.
Numbers that matter — prices, totals, invoice amounts — come from your rules and systems, not from the model's guesswork.
06 /
Hard limits
Some things our systems don't do on their own, no matter how the build is configured.
Never moves money or initiates payments.
Never changes your pricing, margins or terms.
Never sends a quote without your approval.
Never pretends to be a person when a customer sincerely asks.
07 /
Before go-live
Nothing goes live on your real customers until you've seen it work.
Builds are tested on your real (or anonymized) examples in a staging setup first.
We agree success checks, a rollback plan and who to call before launch.
Changes after launch go through the same review — no silent updates.
08 /
If something goes wrong
You can stop any agent, any time, and a person at dev1 owns the problem until it's fixed.
Every build has a pause switch your team can use without calling us.
Exceptions route to a named owner on our side and yours.
If we learn of a security incident affecting your data, we notify you within 72 hours and share what happened and what we changed.
09 /
When you leave
You're not locked in. Your rules, prompts and workflow documentation are yours.
We revoke every credential we were given and confirm it in writing.
You get an export of your configuration, documentation and logs.
Data we hold is returned or deleted within 30 days of offboarding, unless the law says we must keep it.
10 /
Certifications — the honest version
Our hosting, database and model providers hold their own certifications for their services. Those don't automatically certify dev1 or your full build.
dev1's own status: not yet SOC 2 audited. We build on providers that are, including Vercel, Supabase and Anthropic.
If your business needs a specific certification, audit report or contract clause, we'll confirm what we can provide for your engagement before you sign.
This website
What happens to your audit answers.
Audit answers are stored in a database with row-level security on and no public access; only our server can read them.
Your report lives at a private, unguessable link that search engines are told not to index.
The AI that writes your report gets your business answers — not your name, email or phone — and every figure is calculated by our own formulas.
Payments are handled by Stripe. We never see your full card number.
