Changelog

Until today, packages forced a single channel pick: Voice, SMS, or Voice + SMS. That left two real combinations unaddressed: Inbound + SMS (a receptionist that texts back) and Outbound Voice at all (despite outbound shipping in May).

Now the Create / Edit Package screen has three independent toggles:

• Inbound Voice — AI receptionist that answers calls
• Outbound Voice — AI dialer for cold calls, follow-ups, win-back
• SMS — 2-way text + MMS + broadcasts

Outbound voice is treated as a first-class billable channel with its own pricing fields:

• Per-minute rate (often higher than inbound — connection rate + compliance)
• Included minutes / month
• Overage rate per minute
• Daily outbound call cap (optional safety guard)

Existing packages were migrated: 31 single-channel Voice packages became Inbound Voice; the 3 combo packages kept Inbound + SMS. Outbound is opt-in everywhere — no existing client gained a capability they did not already have.

Visit Longneck Luck to claim. The OG tile shows up alongside Giraffe, Whale, Giraffish Millionaire, and $GARY Holder bonuses. Daily reset at 00:00 UTC.

Specifically: every voice agent receives these dynamic variables on each call, localized to the agent's configured timezone:

• {{current_time}} — e.g. "3:47 PM"
• {{current_time_24h}} — e.g. "15:47"
• {{current_day_of_week}} — e.g. "Wednesday"
• {{agent_timezone}} — e.g. "America/Detroit"

Combined with the existing {{today_date}}, agents can now resolve relative dates correctly ("tomorrow morning", "this weekend") and skip windows that have already passed when the call comes in late.

• HVAC, Plumbing, Electrical, Roofing, Pressure Washing, Landscaping, Pool Service, Moving, Window Tint, Med Spa, Chiropractic, Veterinary, Dog Grooming + Boarding
• Each page has its own SEO copy (targets "AI voice agent for X"), industry-specific vocabulary, 6 daily-call scenarios, integration list, comparison table, and FAQ.
• 12 unique demo numbers under the Gary Club agency org — distinct personas, voices, and system prompts. Spanish bilingual on every agent.
• Hub page at gary.club/industries lists all 13 with live-demo badges. Linked from gary.club/hire-an-agent.

Live at docs.gary.club/ai-agency-unlocked/messaging/overview. Covers everything you need to know to register a brand, get a campaign approved, and send compliantly across the US and Canada.

• Outbound SDR (Voice Broadcasts) — dial a contact list with a structured script, capture qualifying outputs, and route warm leads back to your team. Read the docs.
• Outbound Voice Agents — long-running agents that pick up new contacts as they enter the queue (CRM event, webhook, or scheduled CSV pull). One agent per job, separate scripts, separate analytics. Read the docs.
• SMS Agent + Messaging hub are also live for every active agency — Brands, Numbers, Campaigns, and a unified inbox.
• Same fuel pricing as inbound: 25 credits/min ($0.25/min, 30s minimum) per call, 200 credits/mo ($2/mo) per outbound number.
• All outbound calls respect recipient timezone (default 9am-8pm local), DNC suppression, and per-number carrier rate limits.

• Import any recipe directly by URL — no file download needed
• Each recipe includes After Import instructions and a node flow diagram
• Covers SMS auto-reply, missed-call follow-up, deal stage SMS, form-to-contact, outbound AI calls, daily call reports, and more

The org-level toggle on Growth → Messaging Agent now seeds ai_autopilot on every new sms_conversations row. Existing conversations are untouched — your per-conversation overrides still win.

Behind the scenes we wired the WebSocket broadcast at the canonical send site so every outbound (AI, rep, compliance) and every inbound shows up in your inbox the moment it happens. Polling tightened to 5s as a safety net for any event the WebSocket missed.

Open the Growth → Messaging Agent inbox on two devices to see it: a text on one streams into the other instantly.

Every server-side $GARY price read now goes through one canonical oracle (60s fresh cache, Raydium → DexScreener fallback) instead of three independent sources with different update cadences. Your portfolio USD value and the ticker are now always aligned.

The platform completed its Twilio→Telnyx migration on 2026-04-28. Below is what changed for integrators.

New fields on Public API responses

• GET /v1/agents: telnyx_phone_number alongside legacy twilio_phone_number
• GET /v1/agents/:idOrSlug: telnyx_phone_number, telnyx_phone_id alongside legacy twilio_phone_*
• GET /v1/calls + GET /v1/calls/:id: telnyx_call_control_id alongside legacy twilio_call_sid

For new integrations: read telnyx_* first. For existing integrations: continue reading twilio_* for legacy rows; both columns will be present during a 90-day grace window before legacy columns are dropped.

Retired endpoints — 410 Gone

• POST /api/sms/incoming & /api/sms/status — use /api/sms/incoming/:provider and /api/sms/status/:provider (Telnyx).
• POST /api/phone/voice, /call-status, /recording-status, /amd-result, /transcription — Telnyx Call Control replaces these via its own webhooks.
• POST /api/sdr/inbound-voice, /outbound-status, /call-status — same.
• POST /api/phone/token — Twilio Voice JS access token. Click-to-call moved to Telnyx WebRTC; the token is minted client-side now.
• GET /api/phone/recordings/:sid — Twilio recordings archived. Telnyx recordings stream from their own URL on the call row.
• POST /api/phone/tag-team/* — handoff feature pending Telnyx Call Control conference port.

What integrators need to do

• Update field reads to prefer telnyx_*; keep twilio_* as a fallback during the 90-day grace window.
• If you were posting to any of the retired routes, switch to the Telnyx-shaped equivalent OR remove the integration if it was a leftover.

• Theme toggle in the header (sun / moon / monitor; cycles light → dark → system).
• Pre-paint script sets the chosen theme before first render — zero flicker on reload.
• Theme-aware logo: dark "Gary" text on light backgrounds, light "Gary" text on dark.
• 13 new design components for endpoint blocks, code tabs, callouts, property tables, hash-anchored sections, and a theme-aware Scalar reference embed.

The dashboard now treats API keys as a first-class surface:

• Agency sidebar: Team & Trust → API Keys for one-click access.
• Account tab: emerald CTA card linking to /settings/api-keys.
• Per-client settings: list agent_live_ keys minted on this client's behalf, mint new ones with a scope picker + rate-limit + expiry, copy the cleartext once, revoke any time.

Available agency-wide (Settings → Client Portal → sidebar items) and as a per-client override (each client's Portal Visibility section). When enabled, the client portal:

• Surfaces API Keys in the sidebar.
• Unlocks POST /api/client/api-keys so the client can mint agent_live_ keys scoped only to their own data.

• Two key shapes: gc_live_ for agency-wide access, agent_live_ for client-scoped (cross-client reads return 404 — no leakage possible).
• Stripe-Connect-style delegation: agency keys can pass X-Gary-Client-Id per-request to act as one client.
• Outbound webhooks: subscribe to call.received, sms.received, deal.stage_changed, etc. We POST signed JSON to your URL with retries (0s / 30s / 5m / 30m / 4h / dead-letter).
• MCP server: Claude Desktop / Code can use the entire API as native tools — drop a JSON config, restart, done.
• OpenAPI 3.1 spec + Scalar reference UI at docs.gary.club/api/reference.
• The @gary-club/agency Claude Skill ships 12 named workflows (daily triage, pre-call brief, pipeline review, post-call ingest, EOD report, …).

• Open mic — anyone can request stage time, agency admins can lock it.
• Auto-on closed captions in English.
• Mobile-first: stays alive on background, pinned controls.
• Sidebar LIVE pulse + activity feed when the room is hot.

• Middleware now reads agent_id from body OR query, fixing the booking flow.
• New calcom_token_expired notification fires the moment a refresh-token dies — agency owners get an in-app + email heads-up so they can reconnect Cal.com before SDRs hallucinate again.
• connector_events audit-log writes are no longer silently swallowed; failures surface to logs.

• Single channel only — no DMs, no extra channels (locked scope).
• Owner backfill on launch — every existing org owner pre-provisioned.
• New owners + invited members provisioned at creation, not lazily.
• Notification preferences, presence dots, jump-to-bottom, edit indicator, markdown rendering.

• Snapshots UI — point-in-time exports of any CRM view.
• Smart Deal Progression — agencies define stage-advance rules; the engine moves deals automatically.
• Magic Fields — create custom fields from the deal detail page; jsonb under the hood, schema-on-read.
• send_sms automation — automation actions can now send SMS via the configured provider.

• Per-org Telnyx credential storage with AES-256-GCM encryption.
• Per-managed-account inbound webhook with the right Ed25519 public key.
• FUEL-metered send + receive across all per-org numbers.
• Brand-registration FUEL deduct happens on submit, not on approval — so abandoned applications don't cost anything.

• One thread per contact regardless of channel (SMS, email, voicemail).
• Optimistic send — message appears immediately, marked "Sending…" until carrier confirms.
• WebSocket pushes from the backend so other open tabs / clients update live.
• Mobile layout with proper compose bar pinning above the keyboard.

• Five prize tiers from common (1-in-10) to mega (1-in-50,000), weighted with entropy.
• True-odds dynamic blanks so the math is honest.
• NFT prize caps synced (3 / 2 / 1 — Tower / MAGA / Vibin).
• Trash-talk WIN prompt + 250-quip loss pool, all AI-generated.
• Admin claim handler with name + CRM link for prize fulfillment.

• Admin tab at admin.gary.club/admin/giraffes — 4 tabs covering inventory, sales, treasury, and prizes.
• Welcome drop fires on first-time buy only; idempotent.
• Inventory wallet keypair encrypted at rest with WALLET_ENCRYPTION_KEY (perks-treasury pattern).

• Audience engine — compose recipient lists from include + exclude rules across all CRM segments.
• Batch send — Mailgun batch API with adaptive retry; auto-pauses on a complaint-rate spike.
• Test-as picker — search any agency owner to send a test broadcast as them.
• AI composer sees the audience preview, returns 3 subject variants tuned for the segment.

Clicking "Regenerate" on a follow-up email now reveals an input field where you can type specific instructions — for example, "Email Leti not Todd, she referred me" or "Focus on the partnership angle." The AI uses these as override directives when rewriting the email.

This fixes a common issue where the AI misidentified who was spoken to on the call (e.g. emailing the gatekeeper instead of the decision maker) and the only recourse was manual editing.

The live transcript was never loading during calls because the WebSocket connection was missing a required JWT authentication token. The backend immediately rejected the connection with a 4001 Unauthorized error.

Added a /api/phone/ws-token endpoint that generates a short-lived JWT, and updated the frontend WebSocket hook to fetch and append the token before connecting. In-call notes (which send through the same WebSocket) now work as well.

Enrichment Contacts

Enriched contacts now appear immediately in the lead detail panel without requiring a page refresh. Previously, re-enriching a lead would not trigger a contact reload because the enrichment status never visibly changed in local state.

Post-Call Follow-Up Email

The disposition follow-up email now auto-populates with the AI-generated draft, enriched contact email in the TO field, and org signature. Fixed a timing issue where the call ID was lost on disconnect, preventing the analysis fetch. Also increased polling from a single 3-second retry to 5 progressive attempts (up to 25 seconds) to accommodate longer analysis times.

Power Dialer Stats

Call stats (Calls, Connected, Demos, Rate) now update immediately after each disposition with optimistic client-side updates. Fixed a timing issue where the stats sync could miss due to phone re-initialization state transitions.

Inbound SDR — AI Sales Agent

A dedicated AI sales agent that handles your incoming phone calls end-to-end.

Key Features

• 24/7 call answering — Never miss a sales inquiry
• Lead qualification — Discovery questions, pain point identification, real-time scoring
• Dynamic pitching — Uses your pricing packages and business description
• FAQ handling — Custom Q&A entries and knowledge base documents
• Appointment booking — Cal.com and Google Calendar integration with real-time availability
• CRM capture — Saves caller details and conversation summaries
• 7-stage sales playbook — Opening, qualification, discovery, value presentation, objection handling, close, wrap-up
• Customizable — Voice, greeting, system prompt, personality traits, knowledge base

Billing

• 25 credits/minute ($0.25/min) with 30-second minimum
• 200 credits/month ($2.00/mo) per phone number
• Uses shared fuel credits with auto-reload

• 3-tier visibility: system defaults → agency defaults → per-client overrides
• Control sidebar navigation, agent tabs, field visibility/editability, and actions
• Agency Settings → Client Portal Defaults tab for agency-wide config
• Per-client overrides in the Portal & Team tab
• Common presets: managed service (minimal access) or self-service (full access)

• Launch multiple agents per client with location labels
• Clone an existing agent to quickly spin up a new location
• Location switcher dropdown on the agent detail page
• Multi-location badge on the clients list
• Each agent gets its own phone number, call history, and settings