More screenshots

    Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIMuseMuseDotsDots+21 more

    AI Voice Receptionist, Live in 7 Minutes

    1

    Deploy an AI phone receptionist on your own Telnyx and OpenAI accounts: a live number that answers calls, books appointments and logs every call.

    $9.99

    Secure checkout via Stripe

    Money-back guarantee

    Converts to your local currency at checkout

    Security scanned

    AI Voice Receptionist, Live in 7 Minutes

    Example session with this skill installed

    I want to set up an AI phone agent for my small plumbing business. I only need it to answer inbound calls, my call volume and budget are both small, and I don't have a GPU or a domain.

    • Read your context and instructions
    • Compiled the ai voice receptionist

    (From a real test run in a clean session, trimmed of local paths.)

    Path chosen: Quick start (Telnyx to OpenAI Realtime on a small VM). It fits inbound-only, low volume, no GPU and a small budget. With no domain, the server uses an agent.<IP>.sslip.io hostname with an automatic HTTPS certificate.

    Done so far (free, local steps)

    StepResult
    Clone the quick-start build at v1.0.0done
    Create .env from the template; confirm it's git-ignoreddone
    make install (Python 3.12)passed
    make test87 passed
    Pre-flight checklists what's missing below

    Pre-flight

    Settings:
      MISSING OPENAI_API_KEY
      warn OPENAI_WEBHOOK_SECRET is empty: expected until the OpenAI webhook is created in the deploy phase.
      warn TELNYX_API is empty: inbound calls don't need it; check the number in the Telnyx portal.
      warn HUMAN_TRANSFER_NUMBER is empty: the agent cannot hand a caller to a person.
    Git:
      ok   .env is git-ignored
    

    Next, with your approval

    1. Add your OpenAI API key to .env.
    2. Approve the paid steps: a small cloud VM, plus a Telnyx number if you don't have one.
    3. Tell me your business details (name, services, hours, when to hand off to a person) so I can write the persona.

    Then I deploy, set up HTTPS and the OpenAI webhook, point your Telnyx number at it, and run a test call with you.

    Connects securely to your tools. The creator never sees your data.

    About this skill

    In about 7 minutes, get a phone number that people can call and that an AI agent you own will answer.

    You tell your agent "set up an AI phone agent for my business," and this skill walks it through the whole build on your own accounts. That includes choosing the setup, checking your keys, deploying the server, connecting your number, writing your business's script and tuning how the agent sounds. Along the way, your agent stops and asks you before anything that costs money.

    Why use this instead of a hosted voice platform

    • You pay OpenAI's Realtime rate and Telnyx directly, which is cheaper per minute than voice platforms that build on the same model and add their own markup.
    • You also own everything, because the code, the number, the server and the call data all stay in your accounts, so you can change whatever you like.
    • On top of that, you can choose between two builds:
      • The quick start sends calls straight from Telnyx to OpenAI Realtime on a small cloud server, so it doesn't need a GPU.
      • The fully self-hosted build runs your own speech-to-text and voice on a GPU with Claude or a local model as the brain, which can cost less per minute at high call volume and gives you full control of the voice.

    Why use this instead of building it yourself

    Writing the code is the easy part, because most of the time goes into problems that no tutorial mentions. This skill was written from fixing each of the following on real calls:

    • The call connects, but neither side hears anything, even though every health check passes.
    • The agent stops mid-sentence every time the caller says "mhm" or "okay."
    • The agent's answers run so long that callers start talking over them.
    • After an interruption, the agent repeats a lookup it has already done.
    • The line goes silent when the AI model hits a rate limit.
    • The greeting gets cut off, or the call never ends because the agent and the caller keep saying goodbye to each other.

    For each of these, the skill gives your agent the cause, the fix and the log line that shows it's happening, so you don't have to discover them yourself.

    What you end up with

    • You get an AI agent answering a real phone number, and it replies in under a second after the caller finishes speaking.
    • The agent logs every call in Google Sheets, capturing a new caller's name, address and reason for calling and adding notes for returning callers, and it also books and reschedules appointments in Google Calendar.
    • It speaks as your business, using your name, services and hours, and it knows when to hand the caller to a person.
    • If you want outbound calls, the agent can place them as well, and it opens each call with the disclosures the TCPA requires, such as stating that it's an AI and that the call is recorded.
    • You can also turn on live web research during calls, so the agent can confirm an address or an equipment model instead of asking the caller to spell it out.
    • Finally, a scorecard and a tuning playbook help you judge each test call and improve the next one.

    What it costs to run

    You pay your providers directly, and the daily cost depends mostly on how many minutes your agent spends on calls. At current published prices, here is the range:

    • Lower bound, about $5 to $7 a day: the quick start with no GPU, taking about an hour of calls a day. Each minute of conversation costs roughly 8 to 10 cents across OpenAI and Telnyx, and the small cloud server and phone number add about 50 cents a day on top.
    • Upper bound, about $20 to $45 a day: the fully self-hosted build on a GPU that stays on around the clock, with calls coming in all day and night. That covers the GPU at about $5 a day, Telnyx minutes at about $8 a day and the Claude model at roughly $7 to $29 a day, depending on how much of each prompt is cached.

    Most setups, especially small businesses, sit near the lower bound, because a few calls a day costs even less than an hour of conversation. The upper bound is rare unless you are a larger company handling calls for hundreds of customers all day, and at that volume the self-hosted build is also the cheaper choice, since the same round-the-clock calling on the quick start would cost well over $100 a day.

    Why it's $9.99

    Getting this setup right took weeks of building and more than 40 recorded test calls across two working phone agents, and the skill packs all of that into one procedure along with the traps, their fixes and the tuning. As a result, it costs less than an hour of a freelancer's time, and because you buy it once, you can reuse it for every phone agent you build.

    What's inside

    • SKILL.md lays out an eight-phase procedure, and each phase ends with a check before your agent moves on.
    • Eight reference guides cover choosing a build, pre-flight checks, the quick start, the self-hosted GPU build, tunnels, Google tools, outbound compliance and tuning.
    • Two read-only scripts check your keys and settings, and also confirm that your server answers over HTTPS and WebSocket.

    Safety

    • Your agent asks before it spends money or changes any of your accounts.
    • Your keys stay in your own .env file, and the skill never prints or commits them.
    • The bundled scripts only read, use Python's standard library and install nothing.

    The quick start runs on a small cloud server using OpenAI Realtime voices. The self-hosted build runs on a GPU and handles one call per server.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before any payment.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Money-back guarantee

    Doesn't do what its page promises? You get your money back.

    Trust & safety

    Security scan

    Security scan passed 18 days ago

    Requires

    Anthropic
    OpenAI
    Telnyx
    Python 3.11+
    Git
    Google Cloud CLI or Docker
    Telnyx account
    OpenAI API key or Anthropic API key
    Python

    Permissions

    Terminal / Shell
    Network Access
    Read Files
    Write Files
    Environment Variables

    Allowed Hosts

    api.anthropic.com
    api.openai.com
    api.telnyx.com

    Shell: clones the open-source build, installs it in a local virtualenv, runs its tests and the bundled read-only checks. Network: the bundled pre-flight makes read-only API calls to Telnyx, OpenAI and Anthropic to confirm keys; deploy steps talk to the buyer's cloud and carrier only after approval. Environment variables: keys are read from the project's .env and never printed.

    Listed18 days ago
    Updated15 days ago

    What's inside

    Frequently Asked Questions

    The quick start takes about 7 minutes once your accounts and keys are ready, because your agent provisions the server, installs Aria, sets up HTTPS and connects your number for you. The fully self-hosted build takes about an hour on the first run, mostly because the GPU server has to download a large image with the speech and voice models.

    You don't need to write code for the quick start, because your agent runs the commands and explains each step. You do need to be comfortable running an AI agent in a terminal, creating accounts and pasting API keys into a file. The self-hosted build assumes more comfort with servers and Docker.

    For the quick start you need a Telnyx account with a phone number, an OpenAI API key and a small cloud server. You pay those providers directly, which comes to roughly 8 to 10 cents per minute of conversation plus about 50 cents a day for the server and number, so an hour of calls a day costs about $5 to $7 a day. The "What it costs to run" section above shows the full range, and your agent asks before spending anything.

    The skill's reference guides list every failure we hit on real calls, together with its cause, the fix and the log line that shows it. As a result, your agent can usually diagnose a silent call, a cut-off greeting or dead air on its own.

    It suits any business that takes calls to book or qualify work, such as home services, clinics, agencies and sales follow-up. Your agent rewrites the script with your business name, services, hours and handoff rules, and the two example builds are an HVAC intake line and an after-hours sales line.

    No, you don't, because the quick start runs on a small cloud server. You only need a GPU for the fully self-hosted build.

    Yes, it can, and each call opens with the disclosures the TCPA requires, such as stating that it's an AI and that the call is recorded. Before the first outbound call, your agent also walks you through your calling hours and do-not-call list so the setup matches your rules.

    New in Workflow Automation