Telnyx AgentSDK Now Available on Edge Compute Platform

AgentSDK is now available in beta as part of the Telnyx Edge Compute platform. The Agent base class ships in @telnyx/edge-runtime v0.9.2 on npm. Subclass Agent instead of StatefulActor and bring your framework unchanged from LangGraph.js, Vercel AI SDK, or any framework that runs on Node.

What's new

• State management: setState() / getState() with recursive merge and onStateChanged lifecycle hook
• Message history: durable conversation log with adapters for LangChain, OpenAI, and Anthropic formats
• Task queue: queue() for background processing without blocking the request path
• Scheduling: schedule() for delayed execution (reminders, follow-ups, timers). Replaces cron jobs, Celery, BullMQ, and Temporal.
• WebSocket support: webSocket() handler on actor classes for real-time bidirectional communication
• Lifecycle hooks: `initialState, onTask, onConnect
• Platform bindings: zero-credential, zero-hop access to inference on owned GPUs, voice telephony, SMS and WhatsApp messaging, KV storage, shared POSIX filesystem, S3-compatible object storage, and RAG

Why it matters

Every other agent stack has a seam. Your agent runs on a compute platform, its memory lives in a database, its scheduled tasks live in a cron service, its LLM calls go to an inference provider, and its phone calls go through a telephony API. Each is a separate service, a separate hop, a separate vendor taking margin.

Telnyx closes the seam with AgentSDK. Your agent's code, state, memory, model calls, phone calls, and text messages run on one platform, built on carrier-grade infrastructure that Telnyx owns and operates.

What's next

• An AgentClient browser SDK is in progress to bring the same durable state and platform bindings to client-side agents.
• Operating Envelope: an agent governance layer is in design.
• Cross-channel conversation memory is planned to let agents recall context across sessions and channels.

Read the docs at developers.telnyx.com/docs/agent-sdk to get started.