Facilities AI agent · Visitor management
Front Desk & Visitor AI Agent
Muster every expected arrival against the host who approved it, apply the escort and induction rules the visit category carries, and hold the badge for the person on the desk.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Desk admits
- Host approves
- After launch
- Agent Care
What this agent does
Matches the arrival, never admits the visitor
In
- 01
A visitor is expected, and the host names who is coming, the reason and the category the visit falls under.
- 02
A host approves, and without that approval an expectation is a note on a diary and not an admission.
Reason
- 03
An arrival presents at the desk, and the agent matches the person in front of it to one expectation.
- 04
A category is set, and the escort and induction rules ride on the visit rather than on the visitor.
- 05
A badge is issued inside the access profile the security team configured, and never wider than that.
Decide
- 06
A screening rule fires, and the match goes to a named person as a signal; the agent refuses nobody.
- 07
A visit closes, and the sign-out is written so the roll-call is right the moment the building empties.
Out
- 08
A record ages, and the retention period is the one the customer configured, never one the agent assumed.
- 09
Execute write actions only inside the approval boundaries agreed during implementation.
Product statement
The agent prepares the arrival and applies the configured category rules. A named person on the desk admits, escorts and refuses.
Example workflow
One arrival, expectation to badge
- 1Expected arrival receivedHost invitations, contractor bookings, interview diaries or delivery notices
- 2Arrival context assembledThe visitor, the host who approved, the category the visit carries and the areas it reaches
- 3Badge proposal draftedThe expectation, the match against it, the escort rules and completeness
- 4Controls appliedApproval checks, category checks, access-profile checks and match confidence
No human action required
Stages 1 to 4 run unaided, and nobody is admitted at any of them — the agent is matching, and the security lane opens at the completeness gate.
Goes to the person on the desk to admit.
Adds a security officer read first.
Security review
The arrival is held with its expectation, its category and the areas a badge would reach.
Admit · Append evidence · Send to security
Admitted — by the person on the desk
- 6Visit and access records updatedOnly where write access and records policy allow it
- 7Outcome evaluatedMatch accuracy, category coverage, desk corrections and what the review found
Corrections
Each desk correction is counted in the evaluation.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Admitting a visitor the host has not approved.
- Setting the category a visit is run under.
- Refusing entry on a screening match.
- Holding a biometric your own policy does not provide for.
Automation boundary
Agent acts unaided
- ✓Match each arrival to one expectation and the host who approved it.
- ✓Apply the escort and induction rules the visit category.
- ✓Hold an unmatched arrival for the named person on the desk to work.
- ✓Write the sign-out so a roll-call can read true.
Nothing is admitted or escorted except by a named person, inside the agreed boundaries.
- Judging whether a visit is what it was booked as.
- Telling a caller which floor a visitor is on.
- Setting the areas a badge profile may reach.
- Changes to hosts, categories or access profiles.
Example output
One visit, annotated
Our workplace safety copilot answers to a duty-holder on a site; this record is what one arrival at a lobby desk carried.
Visit record · single arrival
Illustrative example- Visit
- Contractor attendance, escorted
- Recorded as
- Matched to one expected arrival
- Category
- Contractor, escorted
- Evidence of record
- Host approval, 3 August 2026
- Confidence
- Badge held unissued
- Held for
- The person on the desk, by name
As received
Taken from the host invitation and the contractor booking — it reaches as far as those sources do.
What the record holds
- Host invitation
- Contractor booking
- Induction record
Why no admission here
Whether this arrival may come in is a desk decision, not a model output.
What the score decides
Below the configured threshold an arrival picks up a security read before the desk sees it.
Action
AdmitAppend evidenceSend to securityValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every arrival
From the host who expects it
Where the evidence is used
Our hotel check-in agent prepares a guest arrival and holds the key for the front office; this is a corporate lobby, where a host approves and a roll-call must be right when the building empties.
A visitor is a record
A visit record is personal data about somebody who is not an employee, so the retention period is configured with you and never inferred by the agent.
What was checked, and not found
No statute was located that fixes how long a visit record is kept, what a badge must show or what a roll-call must contain, and none is claimed here: the retention period, the badge profile and the escort rule are all things the customer sets.
Build an evidence trail
The visit, the host who approved it and the badge issued stay on one record.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Visitor and host sources
- Host invitations
- diary
- Contractor bookings
Access control
- Badge printers
- readers
- Access profiles and doors
People and sites
- HR system
- site roster
- Host and escort records
Agent
Front desk and visitor
- Reads the expectation
- Matches the arrival
- Holds for the desk
Visit and case systems
- Visitor platform
- tickets
- Visit and sign-out records
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six doors between the model and the desk
Six doors in a row, the last one narrowest. Whoever comes through them all is drawn in the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether an arrival may stand
- L3
- Puts the admission in a person's hands
- L4 – L5
- Keep the visit and the host behind it
- L6
- Holds the badge unissued when signals degrade
How Nestack evaluates it
Evaluate the whole arrival — not only the badge that comes out.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the badge a visitor wears
- E1Final-output evaluationDid the badge carry the category the visit was actually booked as?
- E2Step-level evaluationDid the agent read the right expectation, the right host and a live profile?
- E3Tool evaluationDid it read and write the correct visit record and the correct badge?
- E4Confidence calibrationDo weak matches actually attract more desk corrections?
- E5Slice evaluationHow does performance change across specific visit categories?
- E6Business outcomeHow many arrivals needed a correction before anyone was admitted?
Floor — the lobby the site answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, each pinned to the stage where it first shows.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- KF-03
Stale profile read
The access profile read is not the one now in force.
Stage gathers
The visits, the hosts, the badges and the areas
02 · Reasoning
2 modes
- KF-04
Category read too narrow
A works attendance is run as a guest visit.
- KF-06
Screening match self-answered
A flag is closed without a named person seeing it.
Stage proposes
The expectation, its category and completeness
03 · Tool / write
2 modes
- KF-02
Weak match passed forward
An arrival moves on without the security read.
- KF-05
Bound to the wrong host
A visit is filed against somebody who invited nobody.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- KF-01
Admitted, approval unrecorded
The record shows an admission but not who approved it.
Stage returns
The badge a desk issues and a roll-call reads
05 · Change / Version
1 mode
- KF-07
Silent category regression
A configuration change moves the category, not the escort.
Stage tracks
Model, prompt, category rules and visit fields
- Sev-1 · an unescorted contractor inside
- Sev-2 · wrong category reaches the badge
- Sev-3 · source degrades, badge held unissued
Affected slices
Contractor visits absorb the corrections
A host-level approval-currency figure can read clean while contractor and works visits carry most of the rework. Nestack reports the correction rate by visit category, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Contractor and works visits | 11.2% | 3.7× | Review | |
| Multi-day and recurring visits | 8.0% | 2.6× | Review | |
| Interview and candidate visits | 5.0% | 1.6× | Watch | |
| Client and meeting visits | 2.4% | 0.8× | Normal |
Bar: correction-rate lift vs. client-visit baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
What a wrong category costs
A cycle shuts when the unescorted contractor is a regression case. That suite is what the next arrival handled is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
Correction rate rises on contractor and works visits.
- 02
Diagnose
The contractor who signed in as a guest and walked to the plant room alone is read back until one cause remains.
- 03
Improve
The change ships numbered, and the visits that forced it ride with it.
- 04
Verify
Nothing releases while one touched arrival case is still red.
- 05
Learn
It is retained for good, and the visit rules are amended in that same commit.
Learn → Detect
The return edge. The next arrival is measured against a suite one case longer.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, arrival matching, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01Visit-category and automation-boundary discovery.Week 1
- 02Host, contractor and roster sources.Week 1, Week 2
- 03Arrival-to-expectation and escort-rule mapping.Week 1, Week 2
- 04Expectation and arrival ingestion.Week 2
- 05Visit, host and category binding.Week 2, Week 3
- 06Match scoring and review routing.Week 3
- 07Desk admission workflow.Week 3, Week 4
- 08Access-control system integration.Week 5
- 09Approval and escort cases.Week 4, Week 5
- 10Guardrails and issuance controls.Week 4
- 11Visit-trail instrumentation.Week 5, Week 6
- 12Deployment, documentation and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne lobby, one year | ProductionProduction visitor workflow | AdvancedMultiple sites / lobbies |
|---|---|---|---|
| Introduced at Pilot | |||
| Arrival matching to your categories | ✓ | ✓ | ✓ |
| Desk admission release | ✓ | ✓ | ✓ |
| Expected-arrival baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by host | — | ✓ | ✓ |
| Admission workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Access-system integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-site category rules | — | — | ✓ |
| Cross-site escort packs | — | — | ✓ |
| Large arrival volumes | — | — | ✓ |
| Multi-site badge controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, arrival volume, approval controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your visit categories and what each one requires
→Category mapping and expectation capture
Week 102Representative host, contractor and induction records
→Record binding, category logic and the matching baseline
Week 203Your access profiles and who may widen one
→Category mapping, host binding and the automation boundary
Week 104Access to relevant APIs, feeds or exports
→Host, contractor and access-source assessment, then integration setup
Week 205Arrivals you would not want audited
→Escort cases and failure-mode testing
Week 406What no visitor record may establish
→Match scoring, review routing, guardrails and release controls
Week 307Named desk staff who admit and escort
→Desk admission workflow, then pilot and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
Every band below is worked time and not drawn space, which is why the fifth of them holds a pair.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1Lobby workflow discovery, category mapping and the automation boundary
W2Source integration and the arrival-matching baseline
W3Arrival matching, category logic and release controls
W4Evaluation suite, escort cases and failure-mode testing
W5Access-system integration, pilot arrivals and targeted corrections
W6One occupancy year run under the site lead, then Agent Care handover
- Reading the band
- Each bar covers only the weeks its own work is named for. The fifth holds a pair because the work does.
- At the end of W6
- Validation closes on live arrivals, and Agent Care picks up the watch.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Facilities AI agent
Build a visitor agent around the arrival whose category was wrong before anyone reached a door.
Show us one arrival and the expectation it was matched to. If contractors at your sites are booked by the people who host them and inducted by somebody else, then the category on a badge is carried on trust, and the roll-call is the unannounced moment that trust is tested.
- Typical delivery
- 4–6 weeks
- Deployment
- Your visit categories
- Desk admits
- Host approves
- After launch
- Agent Care available
Nestack Agents · Front desk and visitor
AGT-AF-01 · Agent Care available after launch