Sunholo open source

Build AI systems you can understand.

Better boundaries for generated code. Open protocols for connected agents. Two projects you can inspect, run and make your own.

Explore AILANG AI Protocol Platform

An architectural arcade with an orange keystone: clear constraints holding a larger structure together.

AILANG

Programming language · Apache 2.0

Give AI-generated code clearer boundaries.

AILANG is a functional language designed for AI coding agents. Explicit effects describe what a program can do; capabilities constrain its authority; contracts make requirements checkable.

The aim is practical: make generated code easier to inspect, replay and repair.

Try the live demos Get started Source on GitHub

From the entropy presentation

Resolve the choices while change is cheap.

Compare how decisions narrow the possible paths through a program. The AILANG approach moves more of those choices into design, before execution.

Decision Space Explorer showing AILANG effects, contracts and limits narrowing alternative paths before deterministic execution. — Decision Space Explorer, from Programming Languages for AI, not humans. An illustrative model of decision space, not a performance benchmark.

Open full explorer (new tab)

Make effects explicit

File, network and AI operations appear in effect signatures. You can see where a program reaches beyond pure computation.

Check the contract

Express preconditions and postconditions, then use verification and runtime checks to test whether code meets them.

Learn from the trace

Structured execution traces give the coding agent specific evidence to work from when something fails.

See the thinking. Try the code.

View current benchmarks

AILANG development loop presentation: from intent through implementation, verification and feedback.

A development loop built for agents

Follow how intent, implementation and verification fit together.

Explore the development loop

Presentation diagram exploring explicitness, authority and complexity in AI-generated software.

Explicitness without extra guesswork

Explore why visibility and constraints matter when an AI writes the code.

Explore authority and complexity

AI + contractsExtract structured data and check it against contracts. Safe AgentTry tool calling with contract checks on arguments. DocParseExplore document parsing in the browser.

AI Protocol Platform

Application platform · Apache 2.0

An open foundation for connected agents.

Build an agent application around shared protocols, with a web interface and Google ADK orchestration.

The project brings a Next.js frontend, a FastAPI backend and a skills layer together. Explore streaming conversations, generated interfaces and connections to tools and other agents.

Explore the repository Follow the workshop

Connected architectural pavilions illustrating modular services joined through shared interfaces.

Tools and context

MCP connects agents to tools. MCP Apps provides a route to interactive tool interfaces.

Streaming and interfaces

AG-UI carries agent events to the frontend. A2UI describes interfaces the application can render.

Agent collaboration

A2A supports communication between agents; Google ADK provides the orchestration foundation.

Start locally. Understand the pieces.

The workshop offers a local mode that replaces cloud services with local stubs. Follow it to configure model access and run the frontend and backend before setting up GCP infrastructure.

Local mode is for exploration: authentication and persistence are stubbed, and some cloud features are unavailable.

Open the local setup guide

Use it. Question it. Improve it.

Start with a demo or a local checkout. Report a reproducible issue, improve the documentation or contribute a change through the project’s repository.

Sunholo on GitHub Talk to us about your project

Looking for help applying these tools? Explore AI Engineering and Your AI Platform.