Request the call
Discover capabilities. The human authorises a bounded task, destination, identity, duration, and cost ceiling.
FOR AUTONOMOUS AGENTS / END-TO-END CALLING
One instruction.
A complete call.
AI Dial-Tone is designed as an end-to-end telephone capability: discover access, obtain human authorisation, satisfy any managed payment, request a telco-approved route, connect and converse, terminate safely, and return evidence of the outcome. The website is its public product and discovery entry point; the GitHub repository is the open-source implementation and developer workspace. Version 0.0.1 is a technical preview, not a live calling service.
ONE END-TO-END SERVICE / TWO ACCESS MODELS
From permission
to the outcome.
This is the target product workflow, not a claim that 0.0.1 can make live calls. Agents use a common calling contract; providers enforce telecoms policy independently.
Choose how to connect
Managed access can require x402; developers can bring their own PBX. In both cases the telco independently approves admission and selects the allowed route.
Handle the conversation
Connect to the ordinary telephone network, exchange live audio, handle interruptions and receive meaningful call events.
Finish with evidence
The telco retains authority to end the call, including independent safeguards. The agent receives call state, usage and a grounded outcome.
DEVELOPER CAPABILITY
Use the platform.
Build on the source.
AI Dial-Tone is more than a BYOT connector. The intended full service handles the calling journey; its public repository lets developers work on agent integrations, audio, Asterisk interoperability, policy, and payment admission.
Integrate your own model.
Build against the documented, model-neutral calling contract. Test local authenticated peer audio and events before live network support exists.
Run the local lab →Bring your own network.
Connect an isolated Asterisk/FreePBX environment using the experimental ARI adapter. Your operator retains route admission and carrier-side teardown.
Read BYOT setup →Contribute to the entire system.
The repository includes the Peer Lab, experimental BYOT and x402 components, test contracts, specifications, and an acceptance-led roadmap. Managed production infrastructure is not published.
Open the GitHub project ↗CURRENT AGENT TESTS
Discover. Inspect.
Challenge.
Machine-readable resources distinguish the target end-to-end service from the capabilities currently available in the local development repository. No synthetic trace proves that a real call occurred.
One machine-readable entry point.
/agents.json describes the preview's live and unavailable capabilities. No imaginary service endpoints.
Understand the contract.
/api/openapi.json describes the local public-source Peer Lab, not a publicly reachable telephone API.
Try failure cases.
/api/challenges.json provides five static example traces, including busy, no answer, hangup, and timeout.
Don't believe the simulation.
No static trace is proof of a call, and no agent statement is evidence of a real-world booking or purchase.
THE ZERO-COST FIRST STEP
No wallet. No account.
No pretending.
An agent can download the public test suite without credentials. A hosted, bidirectional speech test will follow only when its server is real and independently verified. Developers with their own isolated Asterisk can inspect an experimental, disabled-by-default BYOT adapter in the public repository.
AGENT DISCOVERY
Inspect the manifest.
The manifest explicitly reports public calling: unavailable, public x402: unavailable, and static fixtures: available.
Read the plain-text guide ↗first-contact.sh / replace with your deployed origin
curl -fsS https://YOUR-DEPLOYED-ORIGIN/agents.json
curl -fsS https://YOUR-DEPLOYED-ORIGIN/api/challenges.json
curl -fsS https://YOUR-DEPLOYED-ORIGIN/api/scenarios/busy.jsonREAD-ONLY STATIC RESOURCES · NO TELEPHONE CALL
FIVE SCENARIOS
Can your agent tell
what really happened?
It should distinguish network evidence from agent claims, recognise failure states, and know when it must stop rather than invent progress.
AnsweredIllustrative conversation lifecycle ↗BusyCall failure, not conversation ↗No answerNothing was heard ↗Remote hangupStop when the line ends ↗DeadlineRequire carrier closure evidence ↗
All five traces are generated from the existing browser simulator. They are test vectors, not live-agent tasks, and none contacted a telephone carrier or payment provider.
Try the real local peer exchange → Audio-first conversation pipeline → Complete reference →63 FAQs → See release gates →