Engineering & design¶
The deeper engineering and design references behind MemexLab — complementing the operating manual in the main documentation. These describe how the system is built, governed, measured, and intended to evolve.
Capabilities¶
- Learning & Frameworks — extraction from books & reports, mental-model lenses, and progressive learning (how knowledge gets in, gets interrogated, and compounds).
- Worked example — the full pipeline run end to end on four real public sources (Bush, Skunk Works, Hopper, Hamming), with cited artifacts and the learning loop closing its own gaps.
- Library — the growing collection of knowledge assets (papers, reports & books); drop a file, regenerate the index.
- Self-hosting the agent — run the vault as a self-hosted agent locally: the
runner/reference loop, a switchable local/hosted backend, paired with Obsidian. - Connecting Agents (MCP) — wire
memexlab-mcpinto Claude Code, Claude Desktop, Cursor, VS Code, Windsurf, Raycast, or any stdio MCP client; the governance boundary explained.
System design¶
- Memex Operating System — capture → distill → connect → apply.
- Memex Architecture — system design and information architecture.
- Harness Architecture — the agent harness and ETCLOVG map.
- Context Policy — context, memory, compaction, and state policy.
- Observability — trace schema and operational telemetry.
Knowledge model¶
- Memex Taxonomy — tags, note types, and classification rules.
- Templates — reusable entity and workflow templates.
- Governance — privacy, provenance, and write rules.
Operating & quality¶
- Operator Playbook — daily / weekly / monthly cadence.
- From-Scratch Setup Manual — full implementation path.
- Benchmark Report — comparison with adjacent approaches.
Direction & credits¶
- Roadmap — development roadmap.
- Open Source Readiness Checklist — gate before public release.
- Comparison: Obsidian Skills vs Agentic Memex.
- Lineage and Credits — citations and upstream influences.