Skip to content

Quickstart

About 15 minutes, most of it a one-time dependency install and model download.

Requirements

  • Node.js 26 (the version pinned in .tool-versions) and pnpm (the version pinned in each package's packageManager field; corepack enable pnpm picks it up).
  • Obsidian (optional but recommended for editing the vaults).

Clone and build

git clone https://github.com/tanya-ok/engineering-os.git
cd engineering-os
./scripts/setup.sh   # checks node + pnpm, builds rag/, copies example configs

setup.sh installs and builds rag/, then copies .env.example, rag/vaults.example.json, and rag/routing.example.json to their live counterparts. The first index run downloads the embedding model (intfloat/multilingual-e5-small, quantized, about 120 MB); it is cached under ~/.engineering-os/models.

Index the vaults

node rag/dist/cli.js index --config rag/vaults.json

This indexes the three shipped vault templates (work, ai, user). Point it at your own vaults by editing rag/vaults.json or setting the EOS_*_VAULT_PATH environment variables. Vault paths must sit under the config's allowed_roots (default: this repo), and iCloud paths are always refused - the diode rule that keeps personal data out of a work-side index.

node rag/dist/cli.js serve --config rag/vaults.json   # --port N / --host H to override

curl -s -X POST http://127.0.0.1:8765/search \
  -H 'Content-Type: application/json' \
  -d '{"query": "how do I record an architecture decision", "top": 5, "hybrid": true}'

Request fields: query, top (integer 1-50), vaults, namespaces, hybrid (BM25 + vector via RRF), mmr + mmr_lambda (diversity rerank). See the RAG layer page for the full API.

Make it yours

Open vault-template/ in Obsidian and start replacing the example notes. Point an AI agent at the local search server for grounded answers from your own notes.

Container runtime boundary

Engineering OS runs directly on the host: Node.js serves retrieval and the dashboard, while local files and Beads hold state. No Docker or Podman daemon is required. Keep private vaults and indexes outside the shared container lab.

When reviewing the portfolio, connect to the sibling security lab through its documented local interface. Container workloads use Lima, Kubernetes and containerd; the knowledge layer does not need an additional container or VM.