As a growing consultancy, we had 60+ documents scattered across repos, journals, and local files. Strategy docs in one place, client details in another, process notes in a third. Nothing was searchable, nothing was connected.
Every Claude session burned 10-15K tokens just retrieving context — re-reading files, re-explaining decisions, re-establishing what we were working on. Session continuity was entirely manual. Founders had to re-explain context every single time.
We had plans for overnight agent automation — autonomous agents handling maintenance tasks while we slept. But without a centralized knowledge layer for agents to read from and write to, that automation had no foundation to build on.
We built a two-tier architecture: a GitHub-hosted hub for public knowledge (strategy, processes, project docs) and a local confidential directory for client data and financials. Both indexed by QMD, a local markdown search engine providing instant BM25 + vector search across all documents.
The global CLAUDE.md configuration — previously a sprawling ~270-line file that every session had to parse — shrunk to under 80 lines. Everything else is now retrieved on-demand through search, loaded only when relevant.
Two-tier architecture — public hub on GitHub + local confidential directory
BM25 + vector search — instant retrieval across all documents via QMD
11 overnight agents — memory distillation, quality checks, health monitoring, log rotation
Isolated worktree — agents work on a separate branch, auto-merge to main
Session continuity — current-state.md gives every session full context in seconds
Auto index refresh — search indexes rebuild after every overnight run
We built this because we needed it. Before selling agentic ops to clients, we proved the approach on ourselves — centralized knowledge, autonomous agents, measurable efficiency gains. The same architecture patterns power everything we build for clients.
"We built this for ourselves first. The same patterns and architecture power everything we build for clients. When we say agentic ops works, we're speaking from daily experience — not theory."
— Little Hammer Labs
The same architecture that runs our business can run yours.
Book a Meeting