01 Workflow agents
Cheap and reliable, until they aren’t
Strength. Predictable for known, well defined flows.
Limit. Rigid. Breaks on the first edge case. No judgement, no memory, no learning.
The platform
One place where all of it lives.
The agents, the systems they reach into, the memory they build, the controls you keep, and where the whole thing runs.
Evaluating rather than reading? The full capability list →LayerWhat it holdsState
04Your businessYours
03Your workforceYours
02The control planeOurs
01Models and toolsConfigurable
Isolation · capability based access · shared storage · an audit trail · the primitives every operating system has been built on · built here for a company's AI workforce · isolation · capability based access · shared storage · an audit trail ·
01 · The system
Four layers.Everything your organisation runs passes through the middle one.
Every operating system is built on the same four primitives, and so is this one. Processes are isolated from each other, access runs through typed tools rather than raw credentials, storage is shared so what’s learned outlives any single session, and every action taken is recorded.
Layer 04
Your business
Where the work gets asked for.
Your peopleYour systemsWhere they already work
Layer 03
Your workforce
Everything your agents are, and everything they build for you. Out of the box, built by us with you, or built by your own team inside guardrails you set.
AgentsSkillsMemoryConnectionsAutomationsApplications
Layer 02
The control plane
Every single action passes through here. This is the layer that makes the rest safe.
PermissionsCredentialsApprovalsAuditConfiguration
Layer 01
Models and tools
Configurable by design, with a backup behind the default model, and changing the model touches nothing above it.
Configurable modelsFailover to a backupNative connectors100 plus tools
02 · Positioning
The spectrum of agent capability
Agents aren’t one thing. They sit on a spectrum from rigid task runners, through general assistants, to the category Lifter is built for.
02 Personal assistants
Universal, conversational, broadly capable
Strength. Helpful on any question, with real judgement and reasoning.
Limit. Per person. No shared context, no governance, and no work happens unless someone is typing.
03 Operator agents
General purpose work, with memory and governance
Built for. Agents embedded in the organisation, learning over time, with the controls a real company needs.
The point. A workforce that operates the business alongside your team, whether anyone is watching or not.
03 · The agents
Not a prompt.Six parts your team can read and change.
An agent is a self contained unit with its own identity, its own permissions and its own memory. Not a black box, and not a saved prompt.
TabRiko, Data and ReportingEditable
01Overview name, job title, what it’s responsible forYes
02Skills written procedures, shared across the businessYes
03Connectors the systems it’s joined toYes
04Tools exactly which actions it may takeYes
05Memory what it has learned, in plain filesYes
06Scheduled jobs the work it does when nobody asksYes
Readable
Everything is inspectable
An agent's handbook is written English. Its memory is plain files. Anyone on your team can open either and correct what’s wrong.
Scoped
Permissions per tool
Allow, ask or deny on each individual action inside each connected system, not a switch on the whole application.
Durable
It exists when you aren’t there
Each agent runs as its own process with its own identity. That’s what makes scheduled and event driven work possible at all.
04 · Centralised by design
Set up once.Distribute across the business.Govern in one place.
Connectors, skills, memory, agents, approvals and audit all live in one system. That’s what stops AI adoption fragmenting into forty private accounts nobody can see.
Connectors
Managed in one catalogue. Connect a system once, allocate it to the right teams and agents, revoke instantly.
Skills
A shared library. Pre-built handbooks for common operations, plus your own. Distribute to every relevant agent at once.
Memory
Organisational, not personal. Agents share institutional knowledge rather than one person's chat log.
Governance
One place to set the rules and one place to see what happened.
Every skill written and every lesson learned accrues to the company, not to a single inbox.
That’s the compounding effect, and it’s the reason a business that has run Lifter for a year is in a materially different position from one starting today.
05 · Permissions and approvals
Trust is a dial, not a leap
Every action an agent can take gets one of three stances, set per tool rather than per app. Start narrow. Widen as it earns the room.
Allow
Low risk, reversible, audited. Runs without asking. Drafting a note, building a dashboard, summarising a thread.
Ask
Pauses for a person in the channel your team already uses. The default for anything that touches a customer, costs money or changes inventory.
Deny
Not available to that agent at all. They don’t see it. They can’t try to use it.
ScopeRiko, tool accessEnabled
SYS 01Order management28 / 31
SYS 02Customer support desk19 / 24
SYS 03Email platform12 / 40
SYS 04Finance system0 / 52
Time#cx-approvalsState
14:04Brooke asked to refund $186.50, damaged order, photo on fileAsked
14:09Sam W. approved. Add a $20 store credit too.Approved
14:09Brooke refund processed, credit issued, customer notifiedClosed
Credentials
The model never holds one
Agents work through typed tools, not raw keys. Every credential lives in Lifter's own store and is checked against your policy before any call is made, with secrets redacted on the way back. A prompt injection can raise a request for approval. It can never take an action on its own.
Approvals
An approval is durable state
Requests are persisted, so they survive a restart, expire on a known clock and are recorded with their outcome. Not a message that disappears when a laptop closes, and not addressed to whoever happened to be in the session.
06 · Workflows
Agents and people working the same board.
Work doesn’t belong entirely to the agents or entirely to your team. It moves between them. Columns you define, transitions you type, and a rule on each one saying whether it fires automatically and wakes the next agent, or stops and waits for a person.
Triage Agent
Refund requestOrder 2487, damaged on arrivalREF-481 · just now
Replacement enquirySKU mismatch on a two item orderREF-478 · 3m
Bulk returnWholesale account, 14 unitsREF-476 · 12m
Verify Agent
Eligibility checkedWithin policy, photo on fileREF-475 · 8m
Stock heldReplacement reserved against the returnREF-472 · 14m
Approve Human
$340 refundAbove the agent's limit, waiting on a managerREF-470 · 22m
$50 loyalty creditGoodwill, outside standard policyREF-467 · 31m
Resolve Agent
Refund issuedCustomer notified, logged to auditREF-463 · 1h
Replacement shippedTracking sentREF-460 · 2h
Closed by a personHandled directly, recorded the same wayREF-457 · 3h
Two columns run themselves. One waits for a person. Every card ends up in the same audit trail.
Your columns
Defined by you, not by us
The stages match how your business already works. We don’t ask you to adopt someone else's process to use the platform.
Typed transitions
Each move has a rule
Some fire the moment their condition is met and wake the next agent. Others hold until the accountable person answers. You set which is which.
One board
People and agents in the same place
Nobody is watching a separate queue. A card handled by a person and a card handled by an agent are recorded identically.
07 · Workforce Memory
The longer you use it, the more useful it gets
What your agents know about your business lives in plain files your team can open, read and edit. One person teaches an agent something and every agent benefits. Nothing walks out the door when someone does.
Readable
Plain files
Organised the way a person would organise a wiki. Open it up and change what’s wrong.
Shared
Organisational
Not personal. It stays with the business when a person leaves.
Portable
Survives a model swap
Change the model underneath and the agent keeps everything it learned.
AgeRiko, memorySource
1 moweekly-pack-delivery
Prefers the weekly pack as a document rather than pasted into chat, and wants variance called out before the summary.Taught
Prefers the weekly pack as a document rather than pasted into chat, and wants variance called out before the summary.Taught
3 wkreturns-reporting-basis
Returns are reported net of exchanges. Gross overstates the rate by about two points and should never be used in the performance brief.Taught
Returns are reported net of exchanges. Gross overstates the rate by about two points and should never be used in the performance brief.Taught
2 wkpromotional-week-comparisons
Promotional weeks are never compared against non promotional weeks. Shift to the matching promotional period from last year.Learned
Promotional weeks are never compared against non promotional weeks. Shift to the matching promotional period from last year.Learned
08 · Where it runs
The Gateway runs without the Hub
Hub is optional. On premises, in your own cloud, or air gapped. Pick the deployment that survives your security review.
Hosted
Lifter managed Hub, Gateway in your cloud
The multi tenant operator interface, authentication flows and connector catalogue all live in Hub. Your data stays in your Gateway.
Hub, Lifter cloudGateway, your cloud
Self managed
No Hub, Gateway in your own network
The Gateway loads its own configuration, holds its own credentials and exposes its own tools. Nothing leaves your perimeter.
Hub, optionalNo outbound
Wherever you put the Gateway.We don’t require a phone home.
That’s the one sentence answer to the question a security team always asks, and it isn’t something every platform in this category can give.
09 · Model independence
Lifter is independent of any specific LLM.
Models are configurable by design. Lifter runs on a default model with a backup configured behind it, so if the default has an outage, work fails over to the backup without downtime.
Whichever model is configured
Tool access. The same connectors with the same scopes.
Permissions. Allow, ask and deny apply exactly as you set them.
Credentials. Held by the Gateway. Never delivered to any model.
Approvals. The same queue and the same approvers.
Audit. One ledger covering every action.
What that protects
Agents, skills, memory, connections, automations and built applications sit on Lifter’s layer, along with the platform configuration around them.
Because Lifter isn’t dependent on any specific LLM, a change of model changes none of that. The work you’ve put into your agentic workforce is protected and stays yours.
10 · Takes your brand
Your company's AI operations centre
The Hub is fully white label. Your logo, your colours, your typeface, your domain. To your team it feels like a tool your company built, because to all intents and purposes it is.
Logo and mark
Your wordmark and app icon throughout.
Brand colour
Your accent applied across the interface.
Typography
Choose from a curated typeface set.
Custom domain
ai.yourcompany.com rather than ours.
SettingBrandingState
01Display name Acme OperationsSaved
02Accent colour drives buttons, nav and linksSaved
03Typeface Instrument SansSaved
04Custom domain ai.acme.comLive
Get started
Want to see the platform on your stack?
A thirty minute call. We’ll show you the platform running against systems you recognise, and scope the first agents with you.