maka: Run AI Agents in a Traceable Workspace

Overview

Apache Maka (Incubating) is a TypeScript agent workspace built around a shared Runtime Host and an append-only RuntimeEvent log. Its desktop app runs on macOS, with Windows and Linux support marked as preview; TUI and CLI clients use the same execution authority. Instead of treating a transcript as the only record, Maka keeps events such as model messages, tool calls, permission decisions, and turn endings in the runtime log, which also supports recovery and UI projections.

Maka is aimed at people who want to run and inspect agent tasks on their own machine, including developers evaluating agent harnesses. It does not bundle a shared model account: users connect a cloud API, local model, or compatible gateway. The project publishes evaluation runs using the same model and official verifier, with per-task results in the repository.

Key Features

• Records agent activity as append-only RuntimeEvents, including tool use and permission decisions.
• Provides desktop, TUI, and CLI interfaces backed by one Runtime Host.
• Keeps session data, settings, and run records locally.
• Supports user-provided cloud APIs, local models, and compatible gateways rather than bundling a model account.
• Includes evaluation tooling for experiments, attempts, results, and published per-task reports.
• Offers explicit interrupted-turn resume in Desktop and through the CLI/TUI.
• Provides tools and sandbox boundaries for shell and file operations.

Use Cases

• Developers who want a desktop workspace for assigning tasks to an agent and reviewing its recorded actions.
• Teams comparing agent harness behavior through repeatable evaluations and per-task records.
• Users who want to bring their own model connection and keep workspace records on their machine.
• CLI-oriented developers who need to run a task non-interactively or work through a terminal interface.

Getting Started

For a source checkout, install dependencies and start the desktop app:

git clone https://github.com/apache/maka.git
cd maka
npm ci
npm run dev

You will need a supported Node.js version and an API, local-model, or compatible account connection. See the README for detailed setup, CLI instructions, and release information.

Alternatives

• synara: Synara focuses on coding-agent sessions with project files, terminals, previews, and Git review, rather than Maka's event-logged agent runtime.
• Kun: Kun provides a shared desktop and terminal runtime for agents across multiple kinds of work, while Maka centers on an append-only execution log.
• holaOS: holaOS connects agents with apps, chat tools, files, and integrations, whereas Maka emphasizes recording agent execution and permission decisions.
• sandbase-harness: SandBase Harness is an infrastructure runtime with sandboxed tools, credentials, and an MCP bridge, rather than Maka's desktop agent workspace.

Considerations

Maka is marked as new and has been changing quickly, so interfaces and experimental capabilities may change. Recent commit and release activity is high, and the measured contributor base is broad rather than concentrated around one person; however, the project's short history limits what that activity can establish about long-term stability. Windows and Linux are described as previews, and the current Apache release is incubating. The README also notes that API keys are stored in a local plaintext credential file readable by the operating-system account, and that older JSONL transcripts and Electron safeStorage credentials are not imported during upgrades. Review the security policy and workspace privacy notes before adopting it for sensitive workflows.