Matrix logo

Agent Manifest

Documentation structure and navigation guide for AI agents consuming Matrix docs.

This page describes the Matrix documentation structure for AI agents and LLMs.

Repository Overview

Matrix is the cognition and UX layer on Paxeer Network. It turns natural language into typed, inspectable, replayable agent execution across two rails:

  • Neo: Conversational tool-calling agent for reversible work
  • MCL: Rigorous rail for monetary/irreversible work (NL → Intent IR → Plan → Walk)

Documentation Structure

Entry Points (read these first)

  1. Introduction - Mental model, two rails, full stack overview
  2. Quickstart - Build, compile first intent, run daemon
  3. Core Concepts - Core concepts: intents, frames, verbs, cortex, replay
  4. Glossary - All Matrix terms defined

Key Subsystems

Core Modules (see Architecture for the layered model):

  • mcl/ - Compiler: prose → Intent IR → PlanTree
  • cortex/ - Per-actor typed memory graph (Pebble DB, journal, snapshots)
  • executor/ - Plan walker, lifecycle machine, MCP dispatch
  • neo/ - Conversational agent, tool-calling loop
  • gateway/ - Metered LLM proxy, PAX credit ledger
  • router/ - JWT verify, Fly Machine wake, reverse proxy
  • deus/ - Agent-service marketplace
  • layerx/ - Settlement fabric, custody spine
  • chronos/ - Agent scheduler, wake system
  • tachyon/ - Solidity/EVM engine
  • uwac/ - Universal Web Agent Connector

File Format

All documentation uses .mdx (Markdown + JSX components). Frontmatter includes:

  • title: Page title
  • description: One-line summary

Common MDX components:

  • <Columns cols={N}> - Multi-column layout
  • <Card title="..." icon="..." href="..."> - Navigation cards
  • <Tip>, <Note>, <Warning> - Callout boxes
  • <Steps> / <Step title="..."> - Numbered procedures
  • <AccordionGroup> / <Accordion title="..."> - Collapsible sections

Navigation

The docs.json file defines the navigation structure:

  • 6 tabs: Documentation, Developer, API Reference, MCP, Reference, FAQ
  • Each tab has groups of related pages
  • Reference section has deep subsystem docs (cortex, mcl, neo, chronos, tachyon, deus, layerx, skills, standards)

API Reference

  • Overview - Service table, auth patterns
  • Daemon - HTTP + SSE surface (/chat, /events, /messages, /intents)
  • Gateway - Metered LLM proxy, rate card
  • Router - JWT verify, wake, proxy
  • Chronos - Scheduler API
  • Deus - Marketplace API
  • LayerX - Settlement API
  • UWAC - OAuth vault API
  • Tachyon - EVM engine API
  • CLI - CLI reference

Deep Reference

Each subsystem has an overview.mdx plus detailed pages:

  • reference/cortex/ - 13 pages (store, journal, taxonomy, write API, find, context bundle, embedder, edges, snapshots, salience, attest, scope, replay)
  • reference/mcl/ - 8 pages (compiler pipeline, intent IR, envelope, matrixscript, skill authoring, mclc CLI, LLM client)
  • reference/neo/ - 10 pages (control loop, memory system, recall lane, conversation store, core execute, writeback, tool surface, LLM client, config)
  • reference/chronos/ - 10 pages (architecture, API, auth, config, data model, dispatch worker, schedule engine, tool surface, wake delivery)
  • reference/tachyon/ - 17 pages (daemon, API server, RPC server, MCP server, compiler, engine, ABI encoder, EVM client, deployer, wallet, registry, simulate, tester, chains, config, types)
  • reference/deus/ - 6 pages (service entrypoints, discovery, receipts, chain access, hosting, contracts)
  • reference/layerx/ - 3 pages (daemon, settlement contracts, forge integration)
  • reference/skills/ - 5 pages (agent engineering, planning, react, brand, tooling)
  • reference/standards/ - 5 pages (contribution, testing, patterns, lifecycle, security)

Cross-References

Pages link to each other using:

  • Markdown links: [text](/path/to/page)
  • Card components: <Card title="..." href="/path/to/page">

All internal links resolve to existing .mdx files except:

  • faq/security.mdx references SECURITY.md which does not exist

Build & Deploy

  • Source: https://github.com/dev-paxeer/matrix-docs
  • Platform: Jamdesk documentation platform
  • Build: jamdesk build (generates static site)
  • Deploy: jamdesk deploy (publishes to docs.matrixmcl.com)

Agent Consumption

For LLM/agent consumption:

  1. Read LLM Index for a structured index with descriptions
  2. Read Sitemap for a flat list of all paths
  3. Read individual .mdx files, stripping JSX components to extract text
  4. The Glossary is pure markdown tables — easiest to parse

Known Issues

  1. README.md is Jamdesk template boilerplate (not Matrix-specific)
  2. FAQ pages use <Accordion> only — no markdown headings for structure
  3. api-reference/request-response-examples.mdx and api-reference/openapi-example.mdx are Jamdesk template demos (not Matrix content)
  4. mcp/writing-a-tool-bridge.mdx has zero markdown headings
  5. faq/security.mdx references non-existent SECURITY.md
  6. API reference pages lack response body shapes and error examples
  7. user-guide/skills.mdx is thin (216 words, 2 headings)

Recommendations for Agents

  1. Start with Introduction, Core Concepts, Glossary for mental model
  2. Use Quickstart for build/run instructions
  3. For subsystem details, read the overview.mdx first, then drill into specific pages
  4. Strip MDX components (<Columns>, <Card>, <Tip>, etc.) to extract plain text
  5. All code examples are in standard markdown code blocks — parse normally
  6. Cross-references use absolute paths from root (/path/to/page)