Apache Maka (Incubating)

A high-performance agent workspace that keeps a complete record of everything it did.

An agent harness exists to finish tasks. We hold it to one measure, how many it completes and what that costs, and we publish every run: same model, same official verifier, full per-task record.

Try Desktop Nightly →Build from source

Nightly is a developer build, not an ASF release · Read the architecture

Measured, not claimed. Recorded, not remembered.

Two things the site can prove today: where Maka stands against other harnesses on the same model, and what the runtime actually writes down while it works.

Nine harnesses, one model, the official verifier

Terminal-Bench 2.1 on DeepSeek V4 Flash, every task, scored by the official verifier. The ranking is descriptive; the per-task CSV ships with the report.

Read the report

1Codex

82.0%

2Maka

77.5%

3Pi

74.2%

4DSH

73.0%

5ZCode

70.8%

6Reasonix

67.4%

7OpenCode

65.2%

8Kimi Code

59.6%

9Claude Code

55.1%

pass@1 · reasoning max · cost/pass Maka $0.026

Head to head, same suite

A paired single run against OpenCode on the same tasks. The gap holds up under an exact McNemar test on this suite, and cost per accepted task was at parity.

Read the paired report

+13.5pp · 68.5% vs 55.1%

One Runtime Host

Desktop, TUI, CLI and Eval are thin clients of one execution authority.

How the host works

DesktopTUI / CLIEval

→

Runtime Hostowns execution

The log is the runtime

Every message, tool call, permission decision and termination is an append-only RuntimeEvent. The UI, the next prompt and crash recovery are projections of that log, never the only copy.

Log Is the Runtime

TextFunctionCallpermissionRequestpermissionDecisionFunctionResponseFunctionCallendInvocation