Welcome to Plexium! Some notes to get you started.

[Clarit-AI]

Welcome to Plexium

Plexium gives your repository a persistent, queryable memory layer that both humans and coding agents can use. Instead of re-discovering architecture, module behavior, decisions, and project context from scratch every session, Plexium keeps that knowledge in a repo-local wiki and retrieval layer that can compound over time.

A few useful things to know up front that may not be obvious from a quick skim of the README:

What Plexium Actually Does

Plexium is not just a wiki generator.

It gives you three connected layers:

• a repo-local wiki in .wiki/
• a control/config layer in .plexium/
• a retrieval surface that agents can use through CLI or MCP

That means Plexium can be used as:

• browsable project documentation
• agent-readable project memory
• a retrieval backend for Claude, Codex, and other agent workflows

Important: Plexium Is Per-Repo

Plexium is not a “turn it on once for your whole machine” tool.

You install the plexium binary once, but you run setup inside each repository you want Plexium to manage.

What To Do After Install

Setup wires the system. It does not magically give you a fully rich wiki on day one.

After install/setup, the usual next step is:

plexium convert

That bootstraps real content from the repo into the scaffold.

After that, the best results usually come from having your coding agent refine and expand the wiki, ideally using sub-agents or agent teams for the first larger population pass when your tool supports that.

Core vs Optional Features

The core product does not require a paid model provider.

You can use Plexium for:

• wiki scaffolding
• compile/lint workflows
• retrieval
• MCP-backed agent access

without Ollama or OpenRouter.

Provider setup is for the assistive/autonomous layer:

• background maintenance
• higher-quality enrichment
• model-assisted wiki upkeep

OpenRouter Setup Is Safe

If you want to use OpenRouter, Plexium now supports a safer onboarding path.

You do not need to paste API keys into chat, and you do not need to hand-edit config just to pick a model.

Plexium supports:

• browser OAuth
• safe terminal-native API key paths
• curated model selection during setup
• scripted model selection with --model

So you can use a recommended default, or choose any OpenRouter model you want.

The curated recommendations are there to make the first run easier, not to lock you into a fixed model list.

Curated Models, But Not a Walled Garden

Plexium currently recommends a small curated set of OpenRouter models during setup because that makes onboarding faster and less noisy.

But you are not limited to those.

If you already know the model you want, you can use it directly.

MCP and Retrieval Are First-Class

One thing worth emphasizing: Plexium is not only about generating pages.

It also gives agents a retrieval layer they can use through:

• plexium retrieve
• MCP via plexium pageindex serve
• agent-native integration paths for Claude and Codex

That’s a big part of the value: your agents can query the repo memory instead of just generating it once and ignoring it later.

Memento Is Optional, But Powerful

If you enable Memento, Plexium can treat session history as source material instead of letting important reasoning disappear after the code lands.

That means design rationale, tradeoffs, and discussion context can feed back into the wiki later.

This is optional, but it is one of the most interesting parts of the system.

Temporary Compatibility Shims

Right now Plexium includes temporary compatibility bridges for Claude Code and Codex so Memento works reliably while upstream support catches up.

Those shims are not meant to be permanent product architecture. They are practical compatibility layers so users can benefit from Memento now instead of waiting for upstream fixes.

Secret Safety Matters

One strong recommendation: do not paste secrets into AI chat windows.

That matters in general, and it matters even more in workflows that may preserve or ingest session history later.

Use terminal-native secret flows whenever possible.

If You’re New Here

A good first run looks like:

plexium setup claude
# or
plexium setup codex

plexium verify claude
# or
plexium verify codex

plexium convert

Then start asking the system questions, retrieving from the wiki, and letting your agents help improve it.

If you try it on a real repo, especially one with weak or outdated docs, that’s usually where Plexium becomes much more tangible.

We’d love feedback from early users, especially around:

• onboarding friction
• retrieval quality
• wiki usefulness after first conversion
• Memento workflows
• Claude/Codex integration experience