Every Telnyx build in the open, from a four line SMS send to a durable agent that runs an outage on its own. Start with the newest work on Edge Compute and the Voice AI Agent platform, then take the rest of the library.
A terminal showing the deploy flow: clone the examples repository, add your Telnyx API key as a secret, then run telnyx-edge ship to deploy the example to its own address on telnyxcompute.com. No tunnel and no server to keep running.
The most recent additions on Edge Compute and the Voice AI Agent platform, ordered from the smallest complete app to the ones that run a whole workflow on their own. The rest of the library is one link away.
Newest builds, last 30 days · 531 examples in the full repo
Fetch a URL, summarise it with AI Inference, then cache the result so the second request comes back instantly. The smallest complete Stateful Actor app in the repo, and the one to read first.
Change the model behind a running agent by flipping a KV flag. No redeploy, no restart. Ships with an admin UI that switches the live model between one message and the next.
A LangGraph StateGraph running inside the Agent SDK. Telnyx holds the durable state, the message history and the scheduled work. LangGraph runs the reasoning loop. Bring the framework you already use.
One durable actor per customer, keyed by phone number, alive across days, calls, texts and evictions. The AI Assistant is the voice harness inside it. The agent owns the state. Salesforce included.
Five agents collaborating through one CloudFS filesystem with ordinary POSIX reads and writes, a SQL registry of who did what, and live state pushed to a dashboard over WebSocket.
One durable agent owns an outage end to end. It scores severity with inference, texts affected customers, answers their calls with the incident context, writes the RCA to CloudFS, then schedules its own recurrence check for later.
Answer an inbound call and run the whole speech, inference and speech loop inside a single edge location. Zero credential inference, so there is no model API key anywhere in the code.
Streams live transcription into durable state, one actor per call. On hangup it summarises the transcript, writes it to the actor's own SQL, and texts the summary out. The post-call workflow most teams build by hand.
A voice-first agent whose state outlives the call. LangGraph orchestrates inside the actor, Salesforce holds the CRM record, and the workflow survives reschedules, channel switches and days of silence.
The caller switches language mid-sentence and the agent follows. Five languages behind one number, with no separate speech, model and voice vendors stitched together in application code.
The assistant drives your actual interface. It navigates the app, opens a modal, changes the theme and fills a form by calling JavaScript in the browser during a live conversation.
A HIPAA-aware inbound line that verifies the patient, gives a minimum necessary summary of a normal result, texts a secure portal link, and escalates anything abnormal to clinical staff.