Show HN: Caspian – Talk to Human Tool for AI Agents

Caspian, a communications infrastructure SDK for AI agents, launched on Hacker News on August 21, 2026, with an open-source SDK available on GitHub. The tool aims to give agents a single integration point for talking to humans, abstracting away the complexity of webhooks, message queues, identity management, and channel provisioning.

The project comes from two founders — Dipanshu and Rushant — who described their motivation in the Show HN post: one is a functional programmer, the other has spent time deploying AI employees for companies. They say they observed that 15%+ of issues in the agentic platforms Openclaw and Hermes traced back to communication infrastructure, and they built Caspian to address that directly. The SDK is currently available for Python and TypeScript.

Technically, Caspian is designed as a domain-specific language for communications, drawing heavily on functional programming principles, per the Show HN post. On top of that DSL, the team built a runtime environment that auto-provisions communication channels — which is also their stated monetization path. Users who prefer not to use Caspian's managed provisioning can bring their own tokens instead.

The founders position Caspian against a fragmented landscape of point solutions: single-identity provisioning tools like Agentmail, tool-calling platforms like Composio, and non-agentic support infrastructure like Twilio. Their argument, as laid out in the Show HN, is that all of these are solving pieces of the same underlying communications problem, and Caspian addresses it holistically through an open-source-first SDK. The analogy they use is BetterAuth: rather than rolling your own authentication, you reach for a dedicated SDK — the same thesis applied to agent communications.

The team is soliciting community feedback on two specific questions: how much the interaction channel affects the experience of working with agents, and what features would drive adoption of Caspian. No funding, team size, or customer details were disclosed in the launch post.