/* primo IA · AI cousin */
Open protocols, production implementations, and vertical applications. Three layers, one coherent stack.
Text-only specs and reference implementations. MIT/Apache. The substrate the rest of the stack stands on.
MCP reads. ACP acts. Open protocol for AI agents to control existing application UIs over WebSocket. Reference TypeScript server, conformance suite, npm-published, 5 interactive demos.
Text-only coordination protocol + Python daemon. N parallel agent sessions share a filtered view of each other's actions. No central orchestrator, no tool calls. Multi-provider (Claude, DeepSeek, anything Anthropic-compatible). Stdlib-only, zero deps.
$ pip install primoia-myco
↗
What the open protocols look like when shipped at scale, with a polished UI and commercial support. One product, done well.
Production-grade ACP engine with voice. Multi-platform SDKs for Angular, React, Vue, Flutter, Kotlin, and Swift. Let AI agents see and control your app UI through a structured protocol — no DOM scraping, no vision models, no guessing.
What you can build on top. Each one is its own product with its own audience, but lives in the same family.
Voice-first NFS-e issuer for Brazilian small businesses. Talk to Pepita, she emits the invoice. Built on Vocall + ACP.
AI-curated Hacker News. Scores every Show HN submission across 4 dimensions (Innovation, Technical, Documentation, Utility) with LLMs. ~500 daily submissions classified.
Each layer validates the one above it. Open protocols stay honest because they have to support multiple implementations. Production engines stay grounded because they have to back vertical applications with real users. Verticals stay focused because the plumbing already exists. Most "agent platforms" pick one layer and pretend the others don't matter — this is an attempt at keeping all three in view, even when it means moving slower.
Want to know more about a specific project? Click any card. Or read the about page.