Kiln /

A native Mac app. An open invitation.

From a spark.
To something
real.

Think with models. Build with agents. Understand binaries. Train something of your own.
One shared workshop for the whole process.

macOS · Swift & SwiftUI · Apple silicon recommended
Early development. Source available; no stable release promised.

A sculptural ceramic kiln glowing with molten amber light
Many capabilities. One creative core.

THE IDEA

Your work crosses boundaries.
Your workspace should, too.

Development, analysis, and model experimentation belong in the same conversation. Kiln is exploring a native environment where an agent can help you move between them—with shared services, visible tool activity, and approvals for consequential actions.

INSIDE THE WORKSHOP

Different disciplines.
A common foundation.

01 /

Think & coordinate

Stream conversations across local and hosted providers. Give a team a goal, follow its tasks, and review the results.

Working foundation · model-dependent reliability
02 /

Build & inspect

Use a real terminal, Git tools, file-name search, and MCP connections. Keep the agent close to the work.

Working utilities · full IDE planned
03 /

Understand what’s inside

Import binaries, inspect disassembly and strings, and run analysis tools. Connect Ghidra for deeper recovery.

Native inspection working · Ghidra setup required
04 /

Make models your own

Explore datasets and models, track experiments, and work with Create ML and MLX training engines, including LoRA.

Real training verified · broader GUI QA ongoing

EVIDENCE, NOT PROMISES

Small proofs.
Real work.

5

Provider chat paths

Correct live responses through the app’s session layer from Apple Foundation Models, Ollama, Ollama Cloud, NVIDIA NIM, and OpenRouter.

150

Training iterations

A small transformer trained from scratch, saved its weights, and generated text. A separate real LoRA smoke run saved an adapter.

3

Workers, one artifact

A hosted researcher, implementer, and reviewer completed a real file-summary task with an intact audit chain.

Observed in local QA on November 1, 2026. These examples do not establish production readiness or guarantee every model works. Read the verification record →

BUILD THE NEXT CHAPTER

A workshop gets better
with more hands.

Bring your Swift experience, your sharpest bug report, your real-world workflow, or your eye for good design. Help turn a promising foundation into a tool people can depend on.