I'm developing an open-source project called AkitaLLM.

The idea stemmed from a simple frustration: most LLM tools today try to "think for you." Too much autocomplete, too many implicit decisions, little traceability, and almost no real control.

AkitaLLM goes in the opposite direction.

It's a CLI, local-first tool designed for software engineers that enforces an explicit and auditable workflow for using LLMs. No "just throw the prompt and pray."

The pipeline is rigid and intentional: Analyze → Plan → Execute → Validate

Each step is separate, recorded, and verifiable.

The idea is to use AI as an engineering tool, not as an oracle.

Key points of the project:

Orchestration of LLMs focused on control and predictability

Real difference between analysis, planning, and execution

Integration with AST (Tree-sitter) to truly understand code

Validation through tests, diffs, and rollback

Extensible architecture via plugins

Zero hype, zero "magical AI"

The project is still evolving, but it's already live, versioned, and open to criticism (including harsh criticism).

If you also believe that AI should force you to think better—not think for you—perhaps AkitaLLM makes sense for you.

Repo: https://github.com/KerubinDev/AkitaLLM

PyPI: pip install akitallm

Feedback is welcome.

Especially if it's honest.