Overview
Plan breakdown
AI hosts by plan
Usage
Public demo line
+1 (209) 443-6937 · encrypted caller and conversation analytics
Traction
Calls by day
When people call
Screening levels
Outcomes
Demo calls
Open a call to read what the visitor asked, the AI response, verdict, timing, CNAM, and cost.
| Started | Caller / CNAM | Time | Level | Verdict | Spam | Cost | Details |
|---|---|---|---|---|---|---|---|
Call handling classifier
Seven mutually exclusive policy routes. Sensitivity changes the evidence needed for an interruption—not permission to interrogate.
This matrix reflects the prepared screening policy. Runtime rollout and delivery validation are pending; check deployment evidence before treating these routes as active.
Known-contact policy runs first. Every other call is classified from evidence, risk, and the owner’s Level 1–5 setting.
Policy routes
Known / Direct is a bypass. The other six are sibling screened routes, not steps in a sequence.
Manual owner review has exactly two forms: Live chat for one bounded response, or Whisper audio for a private connect-or-voicemail decision. Plain Review is automated clarification followed by Connect or Voicemail.
Sensitivity levels
Select a level to inspect its evidence bar, finite question budget, routing posture, and calibration example.
Sensitivity × evidence
Classify the strength and interruption value of the story—not the caller’s job title alone. Q counts are caller-screening questions.
| Evidence class | 1 · Open | 2 · Lenient | 3 · Balanced | 4 · Protective | 5 · Strict |
|---|
Questions that may change the route
Resolver order
Decision, delivery, and completion are separate
This prevents a voicemail transition, owner decline, or timeout from leaving stale Connect, mic, or End controls on screen.
Review delivery flows
Each branch exposes only the controls valid for its current delivery state.
Rules that never change with sensitivity
How the fleet is configured
Users
| Name / member | Plan | Storage | Status | Forwarding / relink reminders | Screening | Last app activity | Last incoming call | Calls 7d | Platforms & activity (30d) | Operating cost | Signed up | Calls | Build | Host | Voice | Number | Referral |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Cloudflare storage
Loading storage inventory…
DID inventory
Only Customer allocation rows in pool status can be handed to onboarding. Saving a classification here changes Foyer's inventory database only; it does not buy, release, or reroute a carrier number.
Numbers
| DID | Carrier ownership | Status | Voice runtime | Transition | Classification | Label | Assigned to | |
|---|---|---|---|---|---|---|---|---|
Port status
| DID | Order | LNP status | RDD | Last change | First BulkVS call |
|---|---|---|---|---|---|
Register an admin DID
By design, staff can't see who calls this user — caller numbers, contact lists, transcripts, voicemails, recordings, and tuning-note text are never exposed here. Account activity metadata (latest client/IP), operating-cost allocations, counts, and configuration are visible for support.
LiveKit load — Cloud vs Boston
Provider limits
Numbers from each provider's own usage API (60-second server cache), never local estimates. AWS is read once a day from the free AWS Budgets API (its data lags a few hours); Refresh now re-reads AWS at most once an hour. Warning at 70% of a limit, critical above 90%. The owner gets one email per threshold per period.
| Provider · limit | Status | Used | Limit | % used | Resets |
|---|---|---|---|---|---|
Infrastructure
Numbers marked est. are labeled estimates; unavailable billing stays unavailable. Figures are cached server-side and refreshed daily. Refresh now refetches provider data; dated inventory snapshots retain their observation time.
Production services
API, admin, relay, Supabase, ElevenLabs, R2 and Resend use lightweight read-only checks every five minutes. Unconfigured or read-restricted keys are not counted as downtime. Model providers below are measured only from real screening traffic—this dashboard never buys inference.
| Service | Status | Rolling uptime | Checks | Latency p50 / p95 | Last checked |
|---|---|---|---|---|---|
System reliability
Turn uptime excludes caller aborts. Route availability and SLO exclude routes cancelled because another provider won; a slow hedge loser is a real route miss. Token totals are provider-reported, and usage gaps flag attempts that may have been billed without a final usage frame.
ConversationRelay pipeline
Model routes
| Provider / model | Attempt availability | SLO | Attempts | TTFT p50 / p95 | Total p95 | Errors / slow | Failovers in | Tokens in / out | Usage gaps | Last seen |
|---|---|---|---|---|---|---|---|---|---|---|
AI disagreements
What the owner re-filed a call as, versus the tier the model chose — grouped by model and prompt version.
| Model / prompt version | Calls | Re-filed | Disagreed | Crossed the boundary | Disagreement rate | Avg grade gap |
|---|---|---|---|---|---|---|
Safety stops
Screening stops and speech-guard interventions, grouped by what fired. Content is never shown here.
| What fired | Events | Calls | Trend | Outcomes |
|---|---|---|---|---|
Scenario queue
Calibration-scenario candidates from re-files, submissions and incidents. Redaction is a manual step done outside this console — moving a row to redacted records that a human has already produced the synthetic version; it does not redact anything.
Cost allocation
Operating costs use the provider ledger. DID rent is added as an estimated $1.15/month per owned number, annualized by active number-day; it is not double-booked into the ledger.
Cost over time
Days are America/New_York, not UTC. Days with no events render as $0, never as a gap. The window is rolling (it starts at this time of day N days ago), so its first day is partial and its total matches the cards below for the same window.
Cost accounting
Per-user spend
| Member | Screening LLM | Voice | Telephony | Transcription | Storage | Total (cost) | Legacy usage model (+20%) | |
|---|---|---|---|---|---|---|---|---|
Spam callers
A number is flagged once 3+ distinct users report it — one household never brands a caller. This table backs a future consolidated FCC/FTC complaint export.
| Number | Status | Households | Reports | Auto | First reported | Last reported |
|---|---|---|---|---|---|---|
Groups
A group is a bucket a waitlist row belongs to. Pick one to filter the queue below; tick rows down there and use Assign to group to put people in it. A code that belongs to Alpha testers puts whoever redeems it on the free alpha plan.
Invite codes
The old 8-character codes are retired (Nov 5): Foyer no longer makes or emails them. Codes already handed out still work. Anyone can sign up in the app, and members share their 4-character code.
Waitlist
Code requests from the site and the app, referred-first. Inviting from here is retired (Nov 5): sign-up is open in the app. Click any code to copy it.
| Name | Invite code | Status | Group | Source / CTA | Attribution | Referral | Requested | |||
|---|---|---|---|---|---|---|---|---|---|---|
Referral program
A referral qualifies when the invited friend completes a month of use (real calls handled by Foyer and the app opened on several days) — never on signup. Both sides then get a free month banked. Invites are unlimited and there is no cap on months. The exact rules are shown at the bottom.
Vesting pipeline
Where every referral currently sits. Void rows tripped the fraud floor (same mailbox on both sides, a retired handset, or a VoIP line) — nobody's use of Foyer is affected, only the reward.
Top referrers
Invites are unlimited (owner, 2026-10-17): there are no golden tickets to edit. Admitted counts friends who joined with the member's invite code.
| Member | Direct | Chain | Depth | Vested | Admitted |
|---|---|---|---|---|---|
Feedback
Training review
Curated client-submitted copies only. Accepting a row does not automatically retrain a model.
Billing readiness
Checking billing status…
Apple subscriptions are purchased in the iPhone app. Free beta access never becomes a paid subscription automatically. Extend one member’s free period from their detail page.
Free period
Extend everyone, or one group, by a number of days or to a date. It never shortens anyone’s free time. Members who pay through Apple and complimentary accounts are skipped. Every change is recorded with your name and reason, and members get a “Foyer is free for you until …” notification.