Skip to content
fael.tech
  1. fael.tech
  2. Studies
  3. AIKM

AIKM.

A local knowledge manager so you don't rebuild, project after project, what you've already learned working with AI.

Type
Project
Status
Open source soon
Published

The problem

Working with AI produces a lot of material: prompts, context, skills, agent configurations, decisions, evaluation criteria, failures, and lessons. That material ends up scattered across repositories, chats, local files, IDEs, and each tool's own memory.

The result is rebuilding, in every project, things that were already produced, tested, and used in another one. AIKM starts from two questions: how to stop rebuilding that knowledge, and how to make the result of one run improve the next ones.

Local knowledge

AIKM is a CLI that works on a local workspace. Everything starts with a quick capture that lands in the inbox without any classification. Then each item gets a role (source, knowledge, context, procedure, action, evaluation, or output) and a kind (skill, prompt, checklist, standard...), and only then does it count as active knowledge.

$ aikm init ~/aikm
workspace initialized at "~/aikm"
$ aikm capture --text "Revisar contraste do link no tema escuro antes de publicar"
captured aikm:01M47JE4W14MN5FR2DD20NHMAT -> "~/aikm/inbox/2026-10-06-revisar-contraste-do-link-no-tema-escuro.md"
A new workspace. Everything captured lands in the inbox and only becomes knowledge once it's classified.

Controlled distribution

A skill or prompt that lives in another directory is registered in AIKM, reviewed, and approved. From there, it can be selected for a project and published to it as a copy. The original source stays the reference: AIKM never changes it and flags when a published copy falls behind.

Learning from every run

Storing is only half the job. After using a resource in a project and verifying the result, AIKM helps create a learning record with what was observed, the evidence, the diagnosis, the smallest fix target, what changed, and how it was verified. That record goes back into the collection and becomes context for the next decision.

Private by default

The collection is local. Agents can query it through a read-only MCP server that only returns records marked public or non-sensitive and cleared to leave the machine. Anything private stays local.

AIKM repositoryOpen source soon

Updates

  1. First published.
← All studies