SurfTurf Agent Server / flagship product · v2.1.1 · 2026

A managed team of AI agents that run your portfolio.

Six role-specific agents — each with a working email identity — read your tenant mail, process your owner statements, schedule your vendors, and queue the high-stakes decisions for your review. Multi-tenant by design: every action is tenant-scoped, every agent speaks in your brand voice, every statutory date is anchored to your property's local timezone. Set direction once, watch it apply forever.

For PM firms · landlords · operators Standalone or paired with SurfTurf App Multi-tenant + per-tenant timezone Fair Housing-anchored brand voice Operator-controlled · auto-pause cost cap
ST
SurfTurf
SurfTurf Application
Portfolios, leases, financials, owner statements — the data and operations system of record.
+
AS
SurfTurf · new
SurfTurf Agent Server
A continuously running team of AI agents that operate the portfolio on your behalf.
How it works

A team you'd hire — running on rails.

The Agent Server mirrors the AI Agents configured in the SurfTurf application: same roles, same scope, same standing directives — except now they run continuously, with their own email identities, their own schedules, and their own audit trails.

01 ─ TRIGGER
Work shows up, agents wake.
Concrete example: Tenant emails property@. Inbound classifier triages (spam? phishing? urgent?), the right agent is woken, and the trigger context arrives stamped with your property's local date and time.
02 ─ EXECUTE
Reason, act, route, queue.
Concrete example: Property Agent reads the email (HVAC broken), runs a 5-point self-critique on the draft, routes financial impact to Finance Agent for peer review, sends bid requests to top 3 vendors, queues the spend approval. All logged.
03 ─ FINALIZE
Every run, summarized.
Concrete example: "Property Agent: reviewed 2 HVAC bids, $1,200 + $980. Awaiting your approval. 15 min, $0.04 spent, 78% cache-hit." Timestamp + full audit trail + replayable session.
The roster

Six agents, six inboxes, one orchestrator.

Each agent has a defined role, a working email mailbox, a personal SKILL playbook, and a per-agent memory library of templates, prior decisions, and standing directives you've set.

CM
Conference Mode
ops@
Default mode — ask anything and the right agent responds. Fans out morning work, consolidates evening digests, manages cross-agent priorities.
LE
Leasing Agent
leases@
Full leasing lifecycle — renewals, market rent, escalations, late payments, lease document intelligence. Daily automated scan.
PR
Property Agent
property@
Maintenance & vendor coordination — triages service requests, sources quotes, schedules service, tracks resolution. Daily + monthly cycles.
FI
Finance Agent
finance@
Portfolio-level financial intelligence — reads monthly statements, flags variances, drafts owner reports, reconciles to subledgers.
LG
Legal Agent
legal@
Legal review & compliance — lease clause review, document analysis, tracks insurance, inspections, registrations, and approaching deadlines.
IT
IT Support Agent
it@
System health & data integrity — monitors integrations, surfaces anomalies, admin-only escalations.
The operator dashboard
Single pane of glass. One-click control over everything.
today · 7 runs · 2 pending
Property Agent [email protected]
Replied to 3 inbound from Unit 4B (HVAC); scheduled vendor visit Wed.
Auto-Continue 💬Chat 9am CT
Finance Agent [email protected]
Drafted Sept owner statements for 4 properties; flagged $2.4k variance on 21 Maple.
Auto-Continue 💬Chat Approve & Execute
Leasing Agent [email protected]
Idle · last run 11m ago — quotes received from 2/3 plumbers for 14 Park.
Auto-Continue 💬Chat
Pending decisions 2
!
Approve renewal offer for 21 Maple Unit 2Leasing Agent · drafted from template · awaiting owner Qs
Review
$
$1,200 plumber quote for 14 Park kitchenProperty Agent · over standing-directive threshold
Review
📌 Standing directive saved
"For 14 Park, always confirm with the owner before approving any spend over $500."
applied to · Property Agent · since Mar 12
Two ways to deploy

Standalone, or paired with SurfTurf.

The Agent Server runs on its own — every operator gets the dashboard, agent roster, and chat. Paired with the SurfTurf application, agents gain first-class awareness of properties, leases, vendors, and reports — and the activity surfaces inside the portal views you already use.

A
Mode A
Standalone

Run the agent team on its own. The dashboard is your operating surface — see what every agent did, chat with any of them, approve queued decisions.

  • Live agent dashboard with status + activity feed
  • Direct chat + one-click standing directives
  • Pending-decisions queue + approve / dismiss / snooze
  • Per-agent schedule editor + manual triggers
  • Connect M365, Stripe, Plaid, Slack on demand
B
Mode B · extension
Paired with SurfTurf App

Everything in standalone, plus full data consolidation: agent activity threads into property views, portfolio dashboards, owner portals, and the lease / financial entities you already manage.

  • Agents reason over real properties, units, leases, reports
  • Activity surfaces inside the property + portfolio pages
  • Service-request lifecycle wired to SurfTurf tickets
  • Owner portal sees Finance Agent drafts in-context
  • Same per-tenant scoping; no cross-customer leakage
A day in the life

Morning fan-out → parallel work → evening digest.

The operator's day shifts from triaging-the-inbox to reviewing-the-queue. Most days, the high-stakes decisions are handled before 10am.

07:00
Conference Mode fans out
Morning digest dispatched. Each agent picks up its overnight queue: tenant mail, vendor replies, late-payment flags.
07:00 — 09:30
Parallel agent work
Property Agent triages tenant mail and replies in-thread. Finance Agent reads new statements. Property Agent chases vendor quotes. Legal Agent checks deadlines.
09:30
You arrive
Open the dashboard. 6 runs done, 2 decisions queued, 0 errors. Approve renewal offer; dismiss the false-positive flag. Done in 8 minutes.
17:30
Read evening digest
Conference Mode consolidates the day's runs into a single readable summary. Tomorrow's work is staged. No surprises.
Built-in safety nets

Why a property manager can actually trust this.

Multi-layered guardrails the operator never has to think about — but that prevent the failure modes that make most AI products non-starters in regulated, relationship-driven work.

SAFETY · 01
Send-idempotency
Every outbound email is fingerprinted. The same message can't go out twice within a 6-hour window — even if you double-click "approve."
SAFETY · 02
External-contact approval
First-touch outreach to outside parties always queues for operator approval. Replies in established threads flow within a tight guardrail set.
SAFETY · 03
Pre-finalize self-critique
Every external-recipient draft passes a 5-point check (voice, accuracy, confidentiality, operator-intent, approval-package completeness) before the agent calls finalize. Stops 60–80% of operator-revision loops upstream.
SAFETY · 04
Peer-review router
Drafts touching legal language auto-route to Legal Agent; drafts moving money auto-route to Finance Agent. Best-effort, advisory, never blocks the human queue — just makes sure two agents see anything load-bearing.
SAFETY · 05
Content scanner
Pre-send scan blocks PII (SSN, bank, credit card, DL), secrets/credentials, infrastructure references, prompt-injection echoes, and Fair Housing protected-class language. Per-tenant pattern extension via hot-reload.
SAFETY · 06
Per-tenant timezone
Every agent run is anchored to your property's local date and time — not the platform's default. Statutory deadlines (3-day pay-or-quit, deposit-return, eviction notice) compute against the timezone a court would expect.
SAFETY · 07
Per-tenant scoping
Every action is tenant-scoped. Customer A's directives, audit trails, and dedup windows are completely isolated from Customer B's. Cross-tenant access is a hard-fail at the platform layer.
SAFETY · 08
Mandatory finalize
Every run produces a one-sentence summary, even on "no action." No orphan runs. No silent exits. No mysterious gaps in the trail.
SAFETY · 09
Emergency runbook
A documented operator response path for tenant PII leaks, Fair Housing complaints, statutory-deadline misses, maintenance emergencies, eviction kickoffs, trust-account anomalies, and tenant-boundary breaches. Pause everything in one click.
Quality + cost discipline

Performance you can measure. Cost you can cap.

Operators don't want a black-box AI bill at the end of the month. The Agent Server reports its own performance and stops itself before runaway spend.

QUALITY · 01
Tiered model selection
Light · default · heavy. Routine triage runs cheap; complex orchestration runs strong. Mode + per-tenant config + cost-cap clamp pick the right tier per run.
QUALITY · 02
Auto-pause cost cap
Per-tenant monthly cap with 80%-warning + 100%-pause. The hourly cost monitor pauses the offending tenant's agents — not the platform — so other operators are unaffected.
QUALITY · 03
Cache-hit observability
Every run reports its prompt-cache read rate. Target: >70% across the day. Anomalies surface in the Friday digest so directives that churn the cache get caught early.
QUALITY · 04
Daily + weekly digests
7am CST daily perf summary (anomalies, stalled threads, top approvals) and 4pm Friday weekly digest (per-agent cost, approval-rate, cache-hit, suggested directive tweaks).
QUALITY · 05
Eval harness + replay
Golden-prompt scenarios per agent (must-call-tools, must-not-contain, must-include) run in CI. Failed runs save full session captures so you can replay them tool-call-by-tool-call.
QUALITY · 06
Pause / resume / drain
One click pauses the whole server during deploys or incident response. Graceful drain waits for in-flight runs to finalize before exiting — no half-sent emails, no orphaned approvals.
QUALITY · 07
Knowledge pagination signal
Every knowledge-base search reports the full match count. Agents that hit the result cap on a vague query auto-rephrase before drafting, so customer-facing replies cite the right document instead of the convenient one.
QUALITY · 08
Bounded operational state
Activity log + session captures + state files all prune on a nightly retention cron. The dashboard stays fast at year 3 of operating; Fair-Housing & eviction retention windows are honored separately.
Voice & workflow infrastructure

Agents that sound like you. Workflows you actually use.

A generic AI that produces generic copy isn't a property-management team — it's a chatbot. Every customer-facing word is anchored to your operator-authored content library.

VOICE · 01
Standing directives
Per-agent operator-authored "always" rules ("Always confirm with the owner before approving spend over $500 on 14 Park"). Loaded into the system prompt on every run, edited without restart.
VOICE · 02
Few-shot examples
Calibrated reference outputs per agent (how YOU answer a vacancy inquiry, how YOU draft a late-rent notice). Hand-crafted, the agent reads and matches the pattern.
VOICE · 03
Brand voice guardrails
Per-tenant tone + banned-vocabulary file. PM-flavored: anchored on Fair Housing compliance, with state-specific protected-class additions where needed.
VOICE · 04
Workflow templates
Per-tenant playbooks for routine PM workflows — rent-due reminders, late-notice statutory windows, lease-renewal nudges, annual inspections, deposit-return deadlines.
VOICE · 05
Knowledge TF-IDF index
Semantic search across operator-directives, examples, brand voice, workflows, legal templates, customer references, conflict SOPs. mtime-tracked auto-rebuild on edit.
VOICE · 06
Customer references library
Consent-gated tenant testimonials, owner case studies, property success stories — Fair Housing complaint or eviction context never appears here regardless of consent.
Why this is different

Not a chatbot. Not a copilot. A team. Here's why that matters.

01
Real email identities, not "from a bot"
Tenants reply to property@yourfirm.com like they would any team member. The multi-agent structure is the differentiator — flatten it and we sound like every other AI product.
02
Standing directives that persist
Say something to an agent in chat, click 📌, it's saved to permanent instructions and applied on every future run. Auditable, editable, per-agent.
03
Triggered by the work
Inbound mail, scheduled cron, inter-agent handoffs — agents fire automatically. You never have to "ask" them to start working.
04
Per-tenant timezone-aware
Every agent run is anchored to YOUR property's local clock. A 3-day pay-or-quit posted at 11:59pm in your jurisdiction is timely. Posted on the platform's clock, it's one day late from a court's perspective. We get this right.
05
Self-critique before send
Every external-facing draft is reviewed by the agent itself across 5 axes (voice, accuracy, confidentiality, intent, completeness) before it queues for your approval. Cuts revision-loop traffic 60–80%.
06
For PM firms — not against them
Built to round out the value proposition for management firms, not replace them. Your firm runs leaner; your operators elevate from doing-the-work to setting-direction.
Operator Control — what stays in your hands
  • Every external commitment
  • Every dollar movement
  • Every consequential decision
  • The autonomy dial (Ask · Plan · Auto · Bypass)
  • Per-agent, per-property, anytime
  • Full visibility and audit trail
The agents read, draft, route, and queue. You set the dial — they work inside your guardrails, not around them. Your role elevates from triaging-the-inbox to reviewing-the-queue — that's the whole point.