Maka: keeping a complete task record

Maka is an agent workspace. In the README reviewed here, the desktop, terminal interface and CLI share one Runtime Host. Model messages, tool calls, permission decisions and termination events are appended to an event log. The interface, the next model context and recovery state are derived from that record.

Older tool output can therefore leave the next prompt without disappearing from the log. For a repository investigation, the record can show which tools ran, what they returned and which actions received permission. These are useful review clues; a complete log does not itself establish that the task was completed correctly.

Users supply a cloud API, local model or compatible gateway connection. They add and test it, choose a default model and then start a task. Local sessions and settings do not mean that inputs stay on the machine when a cloud model is used.

Maka remains in Apache incubation, with no Apache release yet according to the reviewed README. Nightly builds target developers and testers; Windows and Linux packages are unsigned previews. They are not presented as production releases, and data formats, commands and experimental capabilities may change.

The live record uses SQLite. Older JSONL conversations and credentials are not automatically imported, so an upgrade can show empty threads and require credentials to be re-entered. Resuming an interrupted turn is off by default; enabling it makes further model calls and uses tokens. A retained log is not a promise to replay every action unchanged.

The retained August 30 weekly snapshot ranks Maka fifth, with 1,918 weekly stars and 4,033 total. That is historical attention, not proof of reliability or adoption. This update draws on newly reviewed documentation and does not backdate current features to that snapshot.