Apache Maka: Local AI Runtime Host for Private LLM Workflows
If you don’t like that when working with an LLM your history, settings, and data are sent to third-party servers, you should take a look at Apache Maka.
This is a local wrapper for AI models that, by default, stores sessions, settings, and execution logs on your device.
You can connect a cloud API, a local model, or a compatible gateway.
Maka records the entire workflow — model responses, tool calls, permission decisions, and task completion.
The desktop app, TUI, and CLI use a single Runtime Host. If the process crashes, the state can be restored from the log.
You can run the Runtime Host on your own computer or server and connect to it remotely.
It’s also worth paying attention to Maka’s architecture. The developers have detailed documentation that breaks down the Runtime Host, state storage, crash recovery, component boundaries, and remote connections.
If you’re interested in how to design an agent wrapper in general, there is a lot of useful information there.
For now, the project is under active development, so the data format, commands, and experimental features may still change.
There are builds for macOS, while Windows and Linux are currently available as previews.
A free open-source project under the Apache 2.0 license.