The repo is the language + the Elixir support. This is the thing I am most interested in other people actually adopting.

## Notes / Ideas

### AI Observations
[noizu-labs-ml/artificial_intelligence](https://github.com/noizu-labs-ml/artificial_intelligence)

Ancient (in ML years) notes on composable AI systems, plus a draft paper.

- [Random AI Musing](https://github.com/noizu-labs-ml/artificial_intelligence/blob/master/README.md) — composable systems, arrow of time / time-decay activators.
- [Dynamic Runtime Model Tuning](https://github.com/noizu-labs-ml/artificial_intelligence/blob/master/paper.md) — extra input nodes mapped onto QLoRA-style adapters, plus dedicated output tokens that hint runtime behavior, so a model (or a system around it) can meta-tweak itself before you collapse back to an uninstrumented net.
- [Roko Coin](https://github.com/noizu-labs-ml/artificial_intelligence/blob/master/roko-coin.md) — tongue-in-cheek: fund / provision compute for an emergent intelligence on a chain. See also [noizu-labs-ml/roko-coin](https://github.com/noizu-labs-ml/roko-coin) and [the-robot-lives/rokos-coin](https://github.com/the-robot-lives/rokos-coin).

## Tools / Apps

### SMAH: Smart as Heck
[noizu-labs-ml/smah](https://github.com/noizu-labs-ml/smah)
@tags `Python`

CLI I wrote to put a model on a Linux box in a useful way. On account creation it records metadata about the system and the user's experience level, then uses multi-pass review/revision (and some other prompting tricks) to raise the quality of the final answer. Sibling: [smahca](https://github.com/noizu-labs-ml/smahca) — code analyzer.

### llm-toolkit
[the-robot-lives/llm-toolkit](https://github.com/the-robot-lives/llm-toolkit)
@tags `TypeScript` `Agents`

The session problem: every coding agent has its own transcript format, and none of them want to talk to each other. This migrates, edits, and **compacts** sessions across frameworks so you can rebase a conversation instead of starting over, or move work from one harness to another without losing the thread.

If you only click one The Robot Lives repo, click this.

### direnv-config (`dc`)
[the-robot-lives/direnv-config](https://github.com/the-robot-lives/direnv-config)
@tags `Rust` `Secrets`

Layered config / secret store. `dc` is how we keep Infisical, direnv, and generated passwords from turning into a pile of exported lies. Agent-safe: you can list, diff, and copy without printing values.

This is unglamorous and I will talk it up anyway. Most “AI infra” dies on secret hygiene.

### claude-assist
[the-robot-lives/claude-assist](https://github.com/the-robot-lives/claude-assist)
@tags `TypeScript`

Claude chat-history viewer / editor / rebase. Predecessor energy for `llm-toolkit`; still useful if you live in Claude transcripts.

## Multi Agent Projects

Older public iterations of a still-closed-source line of work: composing multiple models / agents / channels into virtual work groups.

- [noizu-labs/noizu-teams](https://github.com/noizu-labs/noizu-teams) — more of the architecture written down.
- [noizu-labs/intellect](https://github.com/noizu-labs/intellect) — further along on the implementation.
- [noizu-labs/noizu-collab](https://github.com/noizu-labs/noizu-collab) — multi-agent collaborative environment.

# Cache Management

Hierarchical cache invalidation. The idea is old and still correct: you should be able to bust a *tree* of keys (user → session → fragment) without knowing every leaf, across memcache / Redis / APC / file / whatever.

### Fragmented Keys (PHP)
[noizu/fragmented-keys](https://github.com/noizu/fragmented-keys)
@tags `PHP`

The parent library (2014, still pushed). If you only remember one thing from the personal account, this is it.

### Ports

| Repo | Lang | Notes |
| --- | --- | --- |
| [noizu-labs/fragmented-keys-py](https://github.com/noizu-labs/fragmented-keys-py) | Python | |
| [noizu-labs/fragmented-keys-4java](https://github.com/noizu-labs/fragmented-keys-4java) | Java | |
| [noizu-labs/SwiftFragmentedKeys](https://github.com/noizu-labs/SwiftFragmentedKeys) | Swift | iOS cache providers; WIP-ish. |
| [noizu-labs-scaffolding/fragmented-keys-go](https://github.com/noizu-labs-scaffolding/fragmented-keys-go) | Go | Active. |
| [noizu-labs-scaffolding/fragmented_keys_django](https://github.com/noizu-labs-scaffolding/fragmented_keys_django) | Python | Django instrumentation. |

# Scaffolding

## RuleEngine
[noizu-labs/RuleEngine](https://github.com/noizu-labs/RuleEngine)
@tags `Elixir`

Protocols for DB / config-driven runtime rule engines. Fantastic candidate for hooking GenAI (and other models) into site / monitoring behavior without hard-coding the decision tree.

## Current (wip)
[noizu-labs-scaffolding](https://github.com/noizu-labs-scaffolding)

This org replaced the first generation. **New work belongs here.**

| Repo | What |
| --- | --- |
| [core](https://github.com/noizu-labs-scaffolding/core) | Base records and protocols everything else sits on. Entity-reference protocol lives here. |
| [entities](https://github.com/noizu-labs-scaffolding/entities) | Domain objects + repos. Persistence-specific layers: [ecto](https://github.com/noizu-labs-scaffolding/ecto_entities), [mnesia](https://github.com/noizu-labs-scaffolding/mnesia_entities), [amnesia](https://github.com/noizu-labs-scaffolding/amnesia_entities), [redis](https://github.com/noizu-labs-scaffolding/redis_entities). |
| [services](https://github.com/noizu-labs-scaffolding/services) | Long-lived worker pools with health monitoring baked in. |
| [seed_helper](https://github.com/noizu-labs-scaffolding/seed_helper) | Incremental migration seeds. |
| [smart_token](https://github.com/noizu-labs-scaffolding/smart_token) | Smart tokens. |
| [elixir-framework](https://github.com/noizu-labs-scaffolding/elixir-framework) | Rollup of the above. |
| [elixir_ui](https://github.com/noizu-labs-scaffolding/elixir_ui) | LiveView components / hooks. |
| [ex-infisical](https://github.com/noizu-labs-scaffolding/ex-infisical) | Infisical client. |

## Legacy
[noizu-labs](https://github.com/noizu-labs) — archived, kept for history. [SimplePool](https://github.com/noizu-labs/SimplePool) (14★) is the one people actually starred.

### Advanced Elixir Scaffolding
[noizu-labs/advanced_elixir_scaffolding](https://github.com/noizu-labs/advanced_elixir_scaffolding) · [ElixirCore](https://github.com/noizu-labs/ElixirCore)

What this generation was *for*:

- Domain objects / repos with **multiple persistence layers per module** and fallback (Redis → DB → API).
- Annotation-driven JSON (include / omit by format), security checks, PII stripping from logs.
- Fast switching between refs, database records, and structs via `EntityReferenceProtocol`.

```elixir
defmodule RootLevel.NestedLevel.Image do
  use Noizu.DomainObject
  @vsn 1.0
  @sref "user"
  @persistence_layer :mnesia
  @persistence_layer {:ecto, cascade?: true, fallback_load: true}
  defmodule Entity do
    @universal_identifier true
    Noizu.DomainObject.noizu_entity do
      @permissions {[:view, :index], :unrestricted}
      identifier :integer
      public_field :owner

      @json {:*, :ignore}
      @json {:admin_api, :include}
      public_field :last_login

      @permissions :view, {:restricted, {UserShare, :has_permission?}}
      user_field :bio

      @pii :level_0
      restricted_field :social_security
    end
  end
end