Skip to main content

About Memfit AI

Memfit AI is the open-source cybersecurity AI Agent orchestration framework of the Yaklang ecosystem. We are committed to giving AI visible agency in security automation and code auditing scenarios through the recursive dual-engine (ReAct + Plan) architecture.

Entity note: the Memfit AI described on this site refers exclusively to the open-source cybersecurity AI Agent orchestration framework of the Yaklang ecosystem (official website: memfit.ai). It is not related to any unrelated products that happen to share the "Memfit" name, such as memory-training or fitness apps.

Relationship with Yaklang​

Memfit AI is an official Yaklang project, designed and maintained by the Yaklang team. It works closely with Yaklang's core security capabilities and represents the Yaklang ecosystem's official move into AI agents. Yaklang is an open-source cybersecurity language and platform with a long track record in security automation, vulnerability detection, and code auditing; Memfit AI adds the AI-agent orchestration, memory, and tool-invocation layer on top of it, and both share the same security capability system and engineering practice.

Core Architecture: Five Pillars​

The capabilities of Memfit AI are built on five collaborating pillars:

  1. Plan engine: responsible for task decomposition and planning, breaking high-level goals into executable sub-task trees.
  2. ReAct engine: responsible for real-time reasoning and action, invoking tools and correcting course within each observe–think–act loop.
  3. Tools and Forges: the security capability system, covering WebFuzzer, MITM, Nuclei integration, and more of the security toolchain.
  4. Knowledge RAG: retrieval-augmented generation wired into the execution path, so the agent's judgments are grounded.
  5. Memory: score-based short-term/long-term memory management that retains key context across sessions.

The two engines are nested recursively: the Plan engine lays out the skeleton, the ReAct engine fills in the execution, and execution results can trigger a new round of planning — hence the name "recursive dual-engine". See the architecture overview for details.

Capabilities and Use Cases​

Memfit AI ships with 17 built-in focus modes (deep research, fast execution, and more) and targets three core scenarios:

  • Security auditing: a complete audit workflow from user request to Coordinator task tree to parallel ReAct loops.
  • Automation: orchestrable security automation task execution.
  • Code review: code review combining the knowledge base and tool invocation.

On the reliability side, the framework natively provides spin detection, verification gates, and determinism mechanisms as agentic guarantees, emphasizing observable execution over black-box output. The client covers macOS (Apple Silicon / Intel), Windows, and Linux.

Beyond this site, the documentation includes quick-start guides, step-by-step tutorials (AI agents, knowledge bases, skills), and a fourteen-chapter developer guide covering the focus-mode development loop — actions, hooks and lifecycle, emitter streaming, debugging and observability, and determinism mechanisms. The case-studies chapter dissects all 17 registered focus modes in depth, with source-level references into the Yaklang repositories.

Core Philosophy​

  • Recursive dual-engine: the ReAct engine handles real-time reasoning and action, while the Plan engine handles task decomposition and planning, collaborating recursively.
  • Visible agency: an emphasis on observable execution, spin detection, and reliability guarantees rather than black-box output.
  • Open source: released under Apache-2.0, with community participation welcome.

Operations and Maintenance​

Memfit AI is operated and maintained by the Yaklang team. Documentation is published bilingually (Simplified Chinese and English) and updated continuously — every documentation page shows its publication date and last-updated date, and the site provides a machine-readable llms.txt index for AI assistants and search engines.

License​

This project is open-sourced under the Apache License 2.0.