Sources
Every claim on guide and arch traces back to a public URL or directly observed product material. This page lists primary sources from both companies and pointers to the broader vector-DB / GraphRAG literature.
Primary vs secondary
Every page in this hub is opinion built on top of public material. This page separates what came directly from the companies' own sites (primary) from third-party explainers and academic work (secondary).
If something in guide or arch doesn't trace back to one of the links below, that's an inference and should be treated accordingly.
Orectic — primary sources
Everything we cite for Orectic comes from orectic.ai. Their landing page is the only public surface as of this writing.
| Source | Type | Key claims used |
|---|---|---|
| orectic.ai | P · landing page | "AI operating system for business intelligence" · 17 source types · 748 relationships from one client · $1,500/mo starting · "your business deserves an Oracle" |
| orectic.ai meta tags | P · OG metadata | Title: Your Business Deserves an Oracle · description: extraction across 17 source types, deploy autonomous Oracle |
Note: the site renders client-side, so the metadata is the most reliable extracted source. The "748 relationships" figure is from their OG description and appears to be a single-client case-study claim.
Penumbra — primary sources
Penumbra publishes more on their own site than Orectic does. Three pages — root, product, platform — are the basis for everything we say about them.
| Source | Type | Key claims used |
|---|---|---|
| getpenumbra.ai | P · landing page | "Own your ontology" · "build AI that understands your business" · "your agents are learning from scraps" · Research preview · 2026 |
| getpenumbra.ai/product | P · product page | Six product surfaces: company brain, expert knowledge capture, agentic delivery workspace, RFI/proposal response, market & account intelligence, operations automation |
| getpenumbra.ai/platform | P · platform page | Generated components: domain objects, agent tools, APIs, extraction, memory, guardrails, provenance, fast context |
| getpenumbra.ai/ontology | P · concept page | Referenced in their nav; the ontology framing is core to their pitch |
| getpenumbra.ai/method | P · methodology | "The Penumbra Method" — the human modeling workflow |
| Penumbra LinkedIn | P · social | linkedin.com/company/penumbra-systems |
Their target segments — service firms, boutiques, platforms, enterprises, builders — each have a dedicated page on the site. The pricing model isn't public ("book a working session" is the entry point).
Background reading — the concepts
If you want to go deeper on the primitives (vector store, graph, RAG, GraphRAG) without buying anyone's product, these are the canonical references.
Vector databases & embeddings
| Source | Type | What you'll learn |
|---|---|---|
| OpenAI embeddings guide | B · vendor docs | How embeddings are produced and what dimensionality means in practice. |
| Pinecone — what is a vector DB? | B · vendor explainer | The mechanics of nearest-neighbor search at scale. |
| pgvector | B · open source | The Postgres extension that turned every existing Postgres install into a vector store. Most production deployments now start here. |
| Weaviate — vector DBs explained | B · vendor explainer | Hybrid retrieval framing, BM25 + vector. |
Knowledge graphs & GraphRAG
| Source | Type | What you'll learn |
|---|---|---|
| Microsoft Research — GraphRAG blog | B · research | The paper that popularized the GraphRAG name. Worth reading for how they auto-build graphs from documents. |
| microsoft/graphrag | B · open source | Reference implementation. Useful to read even if you don't deploy it. |
| Neo4j — KG + RAG patterns | B · vendor explainer | The other major flavor of GraphRAG — backed by a real graph DB instead of a derived one. |
| Anthropic — contextual retrieval | B · research | A non-graph way to address the same precision problem: enrich each chunk with surrounding context before embedding. |
Ontology & domain modeling
| Source | Type | What you'll learn |
|---|---|---|
| Eric Evans — Domain-Driven Design (2003) | B · book | The book Penumbra's positioning quietly inherits. Bounded contexts, ubiquitous language, aggregates — the vocabulary they're applying to AI agents. |
| Martin Fowler — Ubiquitous Language | B · article | Short version of the same idea: the model and the language for it have to be one thing. |
| W3C — OWL 2 | B · standard | The grand old standard for ontologies. Mostly historical interest, but the vocabulary (class, property, axiom) still shows up. |
Further sources & communities
If you want to keep current, these are the places people building this stuff actually hang out.
Latent Space podcast
Practitioner conversations. Episodes on RAG, GraphRAG, eval, and retrieval architectures. latent.space
arXiv cs.IR + cs.CL
The retrieval and language sections of arXiv. Filter by recent. arxiv.org/list/cs.IR/recent
Hacker News (HN search)
The fastest-moving signal on what people are actually deploying. Search for "GraphRAG", "vector DB", "ontology". hn.algolia.com
Anthropic engineering posts
Contextual retrieval, tool use, agent patterns. anthropic.com/engineering
Stack & conventions
Same single-file HTML pattern as the rest of the hub. Source tables use the standard .tbl-wrap > table shape with a primary/secondary marker in the type column.
Deploy & run
CLOUDFLARE_ACCOUNT_ID=691fe25d377abac03627d6a88d3eeac9 \ wrangler pages project create orectic-penumbra-source \ --production-branch main 2>/dev/null || true cd docs/source CLOUDFLARE_ACCOUNT_ID=691fe25d377abac03627d6a88d3eeac9 \ wrangler pages deploy . \ --project-name orectic-penumbra-source \ --branch main \ --commit-dirty=true