AKMS¶

Adaptive Knowledge Management System is a family of tools for turning project and domain knowledge into deterministic, inspectable inputs for agents, reviewers, learning workflows, and executable-domain adapters.

The system is deliberately split into narrow responsibilities rather than one omnivorous agent framework:

:material-graph: AKMS Core

Compiles frozen-v2 knowledge nodes into a directed graph, performs ranked and exact task-context resolution, and emits loadouts and manifests.
:material-memory: Failure Memory

Optional project-owned failure/lesson history with deterministic recording, compilation, refresh, provider, and CI workflows.
:material-school: akms-learn

Consumes a graph slice read-only and compiles a validated Learning Source Packet with Markdown, bundle, HTML, notebook, and other exporters.
:material-database-import: Node Generation

Batch Picker and generation tools for turning source collections into validated AKMS nodes, including a NotebookLM CLI path.

Optional companion adapter that bridges computational-mechanics LSP excerpts to the MechDSL executable backend while preserving provenance.

The central boundary¶

flowchart LR
    Sources["Papers, docs, human notes,<br/>agent observations"] --> Gen["Node generation<br/>or project recording"]
    Gen --> Nodes["Validated Markdown nodes<br/>and project-owned registries"]
    Nodes --> Core["AKMS core<br/>graph + task context"]
    Mirrors["Source mirrors<br/>legacy or repo2md"] --> Core
    Routes["Path/symbol route indexes"] --> Core
    Core --> Loadout["Loadout + resolution manifest"]
    Core --> Learn["akms-learn"]
    Learn --> LSP["Learning Source Packet"]
    Core -. optional .-> Runner["staged pipeline runner"]

AKMS core is the knowledge boundary in this diagram. The staged runner is an optional consumer and coordinator. You can use the graph, CLI, exact task resolution, MCP tools, learning compiler, or failure-memory provider without running that pipeline.

Ownership at a glance¶

Artifact	Canonical owner
Global domain nodes	Human-curated vault
Local nodes and experiential overlay	Project repository
Code mirrors	Selected mirror provider, projected into the project
Route indexes	Project repository or its deterministic generator
Loadouts and manifests	Generated AKMS artifacts
Failure-memory registry	Project repository
Learning Source Packets	`akms-learn` output directory
Executable mechanics artifacts	Companion adapter output directory

No component silently promotes project experience into the global vault.

Start here¶

Goal	First page
Build and query a local graph	Core quickstart
Resolve required knowledge from task paths and symbols	Exact task resolution
Generate or refresh source mirrors	Mirror providers
Record and compile project failures/lessons	Failure Memory
Compile learning material	akms-learn getting started
Generate nodes from paper batches	Node Generation
Use the optional agent pipeline	Pipeline runner