apache/maka

What it solves

Apache Maka is a high-performance agent workspace designed to execute, track, and benchmark AI agents. It addresses the need for reliable, reproducible agent workflows by ensuring that every action—from model messages and tool calls to permission decisions—is recorded in an immutable, append-only log. This allows users to audit agent behavior, recover from crashes, and evaluate performance consistently.

How it works

At its core, Maka acts as a runtime host that manages the lifecycle of an agent session. It provides a unified interface across Desktop (Electron), TUI, and CLI environments. The system uses a "RuntimeEvent" log as the single source of truth; the UI and agent logic are projections of this log. Users connect their own models (via API or local providers), and the system handles tool execution, file system interactions, and sandbox boundaries. It also includes an evaluation framework to run experiments and compare agent performance using standardized verifiers.

Who it’s for

This project is for developers and researchers building or testing AI agents who require a transparent, auditable, and local-first environment. It is particularly useful for those who need to benchmark agent performance across different models or tasks while maintaining strict control over their data and credentials.

Highlights

• Immutable Logging: Every agent action is stored in an append-only RuntimeEvent log for full auditability and crash recovery.
• Unified Runtime: A single execution authority powers the Desktop, TUI, and CLI interfaces.
• Local-First Security: Credentials and workspace data remain local to the user's machine, with sensitive keys stored in an OS-protected vault.
• Benchmarking Ready: Includes a built-in evaluation framework to measure agent success rates and costs using official verifiers.