Opinionated, ready-to-run graph cookbooks built on Omnigraph. Each cookbook is a self-contained schema, seed, and query set for a specific use case.
| Cookbook | Status | Description |
|---|---|---|
industry-intel/ |
✅ ready | AI/ML industry intelligence graph |
pharma-intel/ |
✅ ready | Pharma competitive intelligence |
second-brain/ |
✅ ready | Personal life automation graph |
vc-os/ |
✅ ready | Venture-capital operating system |
dev-graph/ |
✅ ready | Software-development work: planning, code, release, ops + governance |
company-context/ |
🚧 planned | Internal decisions, traces, actors, artifacts |
biomed-research/ |
🚧 planned | Biotech & medical research tracking |
competitor-intel/ |
🚧 planned | Competitor launches, pricing, positioning |
The bootstrap skill ships with the industry-intel cookbook it builds from, and installs with the npx skills CLI:
| Skill | Description |
|---|---|
industry-intel/skill |
Bootstrap a new SPIKE graph from scratch — choose demo or custom, elicit domain + sources, adapt schema, research seed content, apply + load |
Install (direct-path form):
npx skills add https://github.com/ModernRelay/omnigraph-cookbooks/tree/main/industry-intel/skillDay-to-day operations are covered by the
omnigraphskill from the engine repo (ModernRelay/omnigraph). It is vendored here at.claude/skills/omnigraph/(Claude Code) and.agents/skills/omnigraph(Codex), pinned to the engine release indeploy/railway/Dockerfile, so both clients pick it up in a checkout with no install step;scripts/sync-skill.shrefreshes it after a pin bump. Other clients:npx skills add ModernRelay/omnigraph@omnigraph. The operating guide and schema-design docs live in that repo and on the docs site.
Typical flow: use the bootstrap skill once to set up a new graph, then the omnigraph skill for day-to-day operations.
omnigraph-cookbooks/
├── README.md CLAUDE.md LICENSE
├── railway.toml deploy/railway/ ← container deploy (Dockerfile, config)
└── <cookbook>/ ← industry-intel, pharma-intel, second-brain, vc-os, dev-graph
├── README.md
├── CLAUDE.md
├── schema.pg
├── cluster.yaml
├── seed.md
├── seed.jsonl
├── omnigraph-config.example.yaml
└── queries/*.gq
The industry-intel cookbook additionally ships the bootstrap skill at
industry-intel/skill/. Each cookbook is fully
self-contained — cd in and follow its README.
- Pick a cookbook.
- Install the OmniGraph v0.11.0 CLI and server — see the OmniGraph repo. No server or object store is needed for the initial cluster apply and seed load.
- Follow the cookbook's Quick Start. Every cookbook is a cluster
directory (aligned to OmniGraph 0.11.0):
omnigraph cluster applycreates the graph and publishes the stored queries;omnigraph-server --cluster .serves them — no object store needed to get started.
OmniGraph 0.11 writes graph format v9 and cannot open a v0.10 (format v6)
graph. A cookbook checkout carries no data of its own — cluster apply plus
the seed load rebuild it — so upgrading one is: install 0.11, delete the
cookbook's graphs/ and __cluster/, run its Quick Start again. A graph that
holds real data is different: cluster-managed roots cannot be upgraded in
place, so export it with the 0.10 CLI, apply the cluster fresh with 0.11 and
load the export, following the engine's
v0.11 upgrade guide.
Stop traffic and keep a verified whole-root backup first; never run mixed
old/new processes against one graph.
The industry-intel/ cookbook uses SPIKE, an opinionated graph modeling lens:
Signal: a dated external fact, movement, or observationPattern: a recurring theme formed, contradicted, or driven by signalsInsight: a synthesized interpretation explaining why a pattern mattersKnowHow: an actionable practice or playbook grounded in the graphElement: a concrete product, framework, company, or concept the signals are about
SPIKE is a cookbook-level convention, not a requirement for every graph in this repo.
Create a new folder, add a schema, seed, queries, and docs. Ship real seed data, not placeholders.