Apache Maka

Overview

Apache Maka (Incubating) is a local-first AI agent workspace that treats agent execution as an auditable, append-only event log. Released in early August 2026 and entering Apache incubation the same month, Maka reached 3,096 GitHub stars in its first week with a weekly gain of 1,313 stars.

Unlike agent frameworks that focus on orchestration or memory, Maka focuses on observability and reproducibility: every model message, tool call, tool result, permission decision, and termination event is recorded as a structured, immutable log entry. This makes Maka uniquely suited for regulated-industry agent deployments where audit trails and reproducible execution are mandatory requirements.

The project is written primarily in TypeScript and positions itself as a foundational layer that other agent frameworks can plug into for logging and audit capabilities — rather than replacing those frameworks entirely.

Features

• Append-only event log: Every agent action recorded immutably for full audit trail
• Local-first architecture: All data stored locally with zero cloud dependencies
• Structured event schema: Model messages, tool calls, tool results, permission decisions, termination events
• Reproducible execution: Logs can be replayed to reconstruct agent state at any point
• Permission decision tracking: Explicit record of every permission grant or denial
• Apache 2.0 license: Suitable for commercial and open-source use
• IDE-agnostic: Works independently of any specific editor or IDE
• Multi-format export: JSON, Markdown, and binary log formats for downstream tooling

Installation

# Via npm
npm install -g @apache/maka

# Or clone directly
git clone https://github.com/apache/maka.git
cd maka
npm install
npm run build

Quick Start

import { MakaWorkspace } from "@apache/maka";

const workspace = new MakaWorkspace({
  path: "./my-agent-logs",
  mode: "append-only"
});

workspace.record({
  type: "tool_call",
  tool: "web_search",
  input: { query: "latest AI agents" },
  timestamp: new Date().toISOString()
});

workspace.record({
  type: "tool_result",
  tool: "web_search",
  output: { results: [...] },
  timestamp: new Date().toISOString()
});

Core Concepts

Event-First Agent Design

Maka inverts the typical agent architecture: instead of an agent runtime that optionally logs, the log is the primary data structure. Every agent interaction produces structured events that can be queried, replayed, and analyzed independently.

The Five Event Types

1. Model Message: A response from the underlying LLM
2. Tool Call: An agent requesting a tool invocation
3. Tool Result: The outcome of a tool execution
4. Permission Decision: A grant or denial of agent action
5. Termination Event: The final state when an agent run ends

Log as Source of Truth

Because the log is append-only and structured, it serves as both the execution record and the system of record. This eliminates the need for separate observability tools — the log contains everything needed to debug, audit, or replay an agent run.

Advanced Features

Replay and Debug

Replay a prior agent run to debug or audit:

import { replay } from "@apache/maka";

const events = await workspace.load({ runId: "run-abc123" });
const replayed = await replay(events, {
  onEvent: (event) => console.log(event.type, event.tool || "")
});

Permission Audit

Extract all permission decisions for compliance review:

const permissions = workspace.query({
  type: "permission_decision",
  from: "2026-08-01",
  to: "2026-08-31"
});

Multi-Agent Coordination

Maka supports multi-workspace configurations for coordinating events across parallel agent runs, enabling visibility into multi-agent system behavior.

Pros

• Unique append-only audit model addresses regulated-industry needs
• Apache 2.0 license for unrestricted commercial use
• Lightweight and fast with local-first design
• TypeScript-native for easy integration with Node.js tooling
• Structured event schema is extensible for custom agent types
• Apache incubation provides governance and community backing

Cons

• Very new project with early-stage API stability
• TypeScript-only (no Python SDK yet)
• Focused narrowly on logging — doesn't provide orchestration or tooling
• Limited ecosystem compared to mature agent frameworks
• Requires integration work to connect with existing agent runtimes

When to Use

Apache Maka is ideal when:

• You need an auditable record of agent execution for compliance
• You're deploying agents in regulated industries (finance, healthcare, legal)
• You want reproducible agent runs for debugging and quality assurance
• You're building a multi-agent system and need a unified logging layer
• You prefer local-first, privacy-preserving tooling

It is not a replacement for agent frameworks like LangGraph, CrewAI, or AutoGen — it complements them by providing the observability layer they lack.

Resources

• GitHub
• Apache Incubator