Plexium gives your repository a persistent, agent-maintained wiki that compounds with every commit.
What Plexium Is
Plexium is a repo memory system for human developers and coding agents. It does not replace your source code or your existing docs. It adds a durable knowledge layer beside them: a .wiki/ vault that agents can read before they work, update after they work, and query later through CLI or MCP instead of rebuilding context from scratch every session.
The short version is simple: LLM agents are good at coding, but they are terrible at remembering. Plexium turns that repeated rediscovery into a shared project memory that can compound over time.
What Happens When You Install It
When you run Plexium in a repository, it creates two project-local surfaces:
.wiki/ becomes the durable knowledge layer: pages, navigation, change log, raw source material, and agent instructions..plexium/ becomes the control plane: config, manifest, templates, reports, integrations, and generated instruction files for agents.
From there, Plexium gives you three ways to use that memory:
- read and maintain it as a browsable wiki in GitHub Wiki or Obsidian
- retrieve answers from it with
plexium retrieve or PageIndex over MCP
- keep it in sync with your code through hooks, CI, and optional background automation
Plexium is also deliberately per repository. You install the plexium binary once on your machine, but you run plexium init or plexium setup <agent> inside each repository you want Plexium to manage. It does not silently apply itself to every repo on your machine.
Core vs optional:
The core product is the wiki, manifest, compile/lint flow, and retrieval CLI. MCP setup, marketplace plugins, daemon automation, Ollama/OpenRouter providers, and Memento ingestion are optional layers you can add when you want more leverage.
How It Works In Practice
The day-to-day loop looks like this:
- A user or agent retrieves context from the wiki instead of starting cold.
- Code changes happen in the normal repo workflow.
- Plexium updates or validates the wiki, navigation, and state manifest.
- Hooks and CI catch drift so the memory stays aligned with the codebase.
That means Plexium is not just a wiki generator. It is a system for keeping project understanding durable, queryable, and enforceable while real work is happening.
Major Capability Layers
Wiki Memory Layer
Plexium stores durable project understanding in .wiki/: architecture pages, module pages, ADRs, concepts, guides, log entries, contradictions, and raw source material. Agents read it before they work and write back to it after they work.
Retrieval Layer
Plexium includes a built-in retrieval engine over the wiki. You can query it directly from the CLI with plexium retrieve "query", or expose the same engine over MCP with plexium pageindex serve so Claude, Codex, or other agents can pull relevant wiki context inside their own sessions. The marketplace/plugin bundles wrap that same retrieval surface instead of inventing a second system.
If you want the raw MCP setup path, use:
plexium pageindex connect claude
plexium pageindex connect codex
If you want Plexium to guide the repo setup and optionally write the native MCP config for you, use:
plexium setup claude --write-config
plexium setup codex --write-config
Automation Layer
Plexium can stay passive, or it can stay alive while you code. Git hooks can enforce that wiki changes accompany source changes. CI can check coverage at PR time. The daemon can watch for stale pages, lint issues, and wiki debt. Claude and Codex integrations keep setup, verification, retrieval, and MCP wiring on a single happy path instead of making users memorize the raw commands.
Assistive Layer
