01
Your data stays yours
We never use your data to train models. Deploy in our cloud in Sydney, in your own AWS, GCP or Azure account, or split between the two. You decide where it lives.
Security and trust
An agent workforce you can put your name on.
Centralised access · approval before anything risky · every action logged
Built so that the person who asks whether this is safe gets a real answer, not a brochure. Your IT lead is going to ask. We want to be the ones who can answer properly.
Working through a review? Every capability, grouped →MetricPositionState
100%Credentials held by Lifter. The model never sees a key.Always
EVERYAction logged with approver, decision and resultExportable
03Deployment options. Our cloud, yours, or hybrid.Your choice
We never use your data to train models · one isolated workspace per customer · retention follows your policy, not ours · exportable to your own audit pipeline · we never use your data to train models ·
01 · The principles
Six commitments we make on day one
Most AI security pages drown in jargon. This one doesn’t. Here’s how we think about it in plain terms, the same way we explain it on a kickoff call.
Worth knowing who wrote them. Our engineering lead has built production systems at enterprise scale and holds a master’s degree in artificial intelligence. None of the commitments below were retro fitted because a customer asked. That’s why they’re in the architecture rather than in a policy document beside it.
02
The model never sees your keys
Logins to your storefront, email platform, support desk and the rest of your stack live with Lifter, encrypted. Agents get the ability to do something. They never see the credential that lets them do it.
03
Risky actions pause for a human
Refunds, customer messages, paid spend, anything that moves money or changes inventory pauses in the channel your team already works in, for someone to approve before the agent acts.
04
Every action is logged
Who asked, what ran, what it touched, who approved it. Exportable in real time to your audit pipeline so finance, IT and the board can see what’s happening without asking.
05
Permissions by person, agent and tool
Your support agent can refund up to fifty dollars without asking. Your trade report agent can’t refund at all. You set the lines, we enforce them.
06
One workspace per customer
Your team's data sits in its own isolated environment with its own storage. Not a shared database, not a logical partition. Real separation.
02 · Credentials
The model never sees a credential.Ever.
Agents see actions, not API keys. The Gateway holds the credentials, looks up the right one per call, checks the rules, and never lets a secret reach the model. Here’s the whole path a refund takes.
StepWhat happensSees the key
01Agent wants to issue a refund. Sees the action's name, nothing more.No
02Gateway looks up the right credential. Checks the rules. Auto approve, ask, or block.Yes
03Human approves in the team's channel, with full context, recorded for audit.No
04Gateway performs the action against the vendor on the agent's behalf.Yes
05Agent gets the result. No credential ever exposed to the model.No
Locked down
Secrets stay with us
Encrypted at rest. Resolved in memory at call time. Never delivered to the model, never logged in clear.
Configurable
Different rules per team
Different stances for different agents, different accounts and different value thresholds. You set them, we enforce them.
Auditable
Every action traceable
Who did what, when, and under which rule, for every call. Recorded and exportable to your audit pipeline.
03 · The control surface
Allow. Ask. Deny.
Every tool an agent can use has one of three settings, and you decide which, per person and per role. Not per application. Per tool inside each application, so an agent can read orders without being able to refund them.
Allow
Safe, reversible work. The agent runs without interrupting anyone. Drafting an email, building a dashboard, summarising a thread.
Ask
Anything that touches a customer, costs money or changes inventory. The agent pauses in the channel your team already uses, for a human to approve, deny, or ask for context first.
Deny
The tool isn’t available to that agent at all. They don’t see it. They can’t try to use it.
WhoA typical setupStance
CXHandles refunds up to $50 on its own. Above that, it asks.Allow to a limit
MKTGDrafts campaigns freely, but waits for approval before any send.Ask on send
DATACan read every system in your stack. Can’t write to any of them.Read only
04 · Where your data lives
Pick the deployment that fits your posture
Three options. Pick the one that matches the conversation you would have with your IT lead. The runtime is self contained either way, with local configuration and local credential storage.
Option 01
Our cloud
Hosted in Sydney, managed and monitored by us. For teams who want it running without a procurement cycle. The fastest path to value.
Option 02
Your cloud
Runs entirely inside your own AWS, GCP or Azure account. Your data and your customers' information never leave your perimeter. Right for regulated industries or strict data residency requirements.
Option 03
Hybrid
The admin interface lives in our cloud so we can keep shipping product improvements. The runtime, meaning credentials, tool calls and customer data, runs in your account. A practical middle ground.
We don’t require a phone home.
The Gateway loads its own configuration, holds its own credentials and exposes its own tools. On premises or air gapped, nothing leaves your perimeter. When a security team asks where the data goes, the answer is one sentence long.
A model outage doesn’t become your downtime. Lifter runs on a default model with a backup configured behind it. If the default goes down, work fails over to the backup and your agents keep going, with their memory, skills, permissions and audit trail exactly as they were.
Agnostic to any specific model. Models are configurable by design, and nothing you’ve built lives inside one. Agents, skills, memory, connections, automations and applications sit on the platform, so the work you’ve put into your workforce is protected whichever model runs it.
05 · The audit trail
Every action, every approval, in one place
The log covers far more than outcomes and errors. It’s the full picture of what your AI workforce did, who approved it and why.
FieldWhat we capture, for every actionAlways
WHOWhich agent, which session, which person kicked it offYes
WHATThe action, and the system it touchedYes
WHENPrecise timestampYes
RULEAllow, ask or deny, and who approved it if it was askYes
RESULTSuccess or failure, with secrets redacted on the way outYes
Where it goes
Your pipeline, in real time
Into the audit pipeline you already use. S3, Datadog, Splunk, or the SIEM of your choice. We don’t gate the data behind our interface.
How long it stays
Your retention, not ours
If your environment requires seven years, we keep it seven years. If ninety days is enough, we keep it ninety days. Retention follows your policy.
Finance, IT and the board can see what’s happening without asking anyone.
06 · The ownership promise
What you build here belongs to you.Always.
Most AI platforms keep your configurations and learnings locked inside their system. We’re built on the opposite premise. The agents, skills, memory, connections, automations and applications you develop are yours to do with as you choose, and off boarding is supported rather than obstructed.
Agents
Fully exportable
Every agent you build or we build with you, including its handbook and configuration.
Memory
Portable and readable
Plain files. Yours to take, and readable without our software.
Contract
No lock in
Monthly subscriptions can be cancelled at the end of any billing cycle.
Models
Configurable by design
Agnostic to any specific LLM, with a backup behind the default. Changing the model changes nothing you’ve built.
Brand
Your identity, your domain
Deployed under your brand, on your own domain, with us quietly in the corner.
Exit
Off boarding supported
If you leave, we help you leave properly. Obstructing that would say more about us than it would about you.
Want the longer write up?
Send us your security questionnaire
Email security@lifter.work and we’ll reply within one business day. If you need a call with our engineering lead, just ask in the email.
We would rather answer the hard questions early than have them surface halfway through a procurement cycle.