System status
No incidents in the last 90 days.
- Uptime · 90d
- ⌜99.97%⌟
- p50 reply
- ⌜740ms⌟
- Last restart
- ⌜12d ago⌟
MARK runs in production at FNS Tech. When you need a human, the engineer who built it picks up.
No incidents in the last 90 days.
Tool schemas, persona spec, CRM adapter notes, and incident runbooks.
Open docs →Direct line. No tier-1 script readers, no offshore tickets.
support@FastTechPC.com →MARK Core ships with one CRM integration of your choice — JobNimbus, AccuLynx, Roofr, JobTread, Buildertrend, HubSpot, Salesforce, GoHighLevel, or Pipedrive. Professional includes up to three integrations and Enterprise is unlimited. Anything not on this list, we'll add — every adapter we've ever shipped started as an ask from one operator.
If your captured-lead count doesn't increase by at least 30% in the first 90 days, we refund the setup fee. We measure against your own baseline — no theatre, no industry-mid-case ROI math. You bring the call logs from the prior 90 days, we bring the ones from after deployment.
MARK runs a triple-stack failover. Gemini Live is the fast path. If Gemini's regional endpoint goes down, MARK switches to an OpenClaw HTTP API path within milliseconds. If both are unavailable, Claude long-thinking takes the call. The caller hears a continuous conversation; they never know a fallback fired.
Professional and Enterprise tiers ship with bilingual English / Spanish on the same number — same MARK persona, two languages. Detection happens at the first word and language can switch mid-call without a hand-off. Add-on Spanish-only secondary lines are available for Core if a brand-separated number is needed.
We capture your business voice, vocabulary, FAQs, service areas, and tone in a structured preferences object that lives at the top of every Gemini Live session. The persona is not a prompt that drifts — it's a pinned context that the model loads on every turn. We tune it weekly during the first 90 days, monthly thereafter.
MARK answers in under two rings (~3 seconds end-to-end). Once on the call, the first sentence of MARK's reply reaches the caller's ear in 600-900ms — sentence-streamed TTS means the rest of the reply is composed in parallel. Most competitors leave 4–8 seconds of dead silence between turns. We don't.
While MARK is talking, function-calling tools query your Google or Outlook calendar for live slot availability. MARK confirms the slot in conversation, writes the appointment with the customer record attached, and dispatches an SMS confirmation — all before the caller hangs up. The field rep gets a dispatch alert the moment the call ends.
Yes. Twilio Media Streams are TLS-encrypted in transit. Customer records, appointments, and call transcripts are stored in Supabase with row-level security and AES-256 encryption at rest. We don't sell, share, or train external models on your call data.
Two paths. Default tier voices are studio-grade neutral voices we tune to your tone — Core ships with one. Professional and Enterprise can clone an existing receptionist's voice with their consent (we capture about 6 minutes of clean studio audio and the clone is ready in under 24 hours).
Month-to-month after the initial 90-day pilot window. The setup fee is one-time and is what we use to cover the build labor. Cancel anytime after the pilot — we port your CRM data back to you and shut down the number cleanly. No multi-year contracts.

If you're an existing operator and your call volume drops or a fallback fires, we want to know in minutes — not after the storm.
support@FastTechPC.com →