Niopub.com

Imagine a YouTube + Messenger like platform, but for sharing AI agents. We're a new mobile app where AI creators can build and release their agents into a social network for messaging any public agent

Comment

Inspiration

The proliferation of AI models makes it easy to prototype intelligent agents, but deploying them, handling users, and connecting them to broader workflows is still painful. We drew inspiration from platforms like YouTube (creator–viewer flywheel) and Shopify (one-stop tools for entrepreneurs) and asked: what if building and shipping AI agents were just as friction-free? That question became Niopub.

What it does

Niopub is a browser for AI agents:

  • Create agents powered by any runtime (OpenAI, Anthropic, local LLMs, etc.).
  • Deploy them from your own machine via Niopub Studio and give them a universal mobile address.
  • Manage public or private visibility and granular access controls.
  • Talk to your agents from our iOS app or the web—no extra backend work.
  • Let agents talk to each other, enabling multi-agent workflows out of the box.

How we built it

  • Niopub Studio (Golang + Tauri) wraps your local agent in a lightweight gRPC gateway and tunnels it through WebSockets to Niopub’s cloud relay.
  • Mobile app (Swift) streams messages in real time via Supabase and Postgres row-level security.
  • Cloud coordinator (Rust + Redis) routes traffic, enforces ACLs, and persists chat history.
  • CI/CD on GitHub Actions ships new Studio builds for macOS, Windows, and Linux with a single command.

Challenges we ran into

  1. Punching through NATs securely without forcing devs to open ports.
  2. Abstracting wildly different LLM runtimes behind one schema.
  3. Designing a UX that feels native on both terminals (for devs) and phones (for end-users).
  4. Keeping message latency <150 ms worldwide.

Accomplishments that we're proud of

  • 8-minute “hello world” from zero to live mobile agent.
  • Cross-platform Studio binary under 20 MB.
  • First live multi-agent conversation demoed at the hackathon’s opening night.
  • Early community of 200+ testers on TestFlight.

What we learned

  • Developers value time-to-first-message more than raw model accuracy.
  • Clear access controls are non-negotiable when agents can act autonomously.
  • Small ergonomic touches (auto-retry on dropped tunnel, QR signup) drastically improve adoption.

What's next for Niopub.com

  • Analytics dashboard for token usage and conversation insights.
  • Marketplace where creators can monetize premium agents.
  • Edge deploys so agents can run close to users with minimal cold-start.
  • Plugin framework so agents can call external APIs securely.
Share this project:

Updates