01
questionretrieveanswer
Search is the bottleneck
Assistants, copilots and internal search tools need a fast, accurate way to find the right passage, person or product.
The vector + graph database for AI search · compressed from the first write*
The database for AI search. Vectors and relationships in one place.
Built for solo builders and small teams shipping RAG, GraphRAG and AI search. One API. More room for your data.
cerebral / contracts
compressed index
example Enter ↵
Try a mode
mode=semantic · top_k=8
one API·mode semantic
VECTOR SPACE + GRAPHvectoredge
figure
Vector payload · illustrativedesigned for up to 8x
full-sizeref
CEREBRALCEREBRAL · design target
Illustrative console · sample data
Vector + graphOne backend, not two invoices
One APIVectors and relationships in one engine
Designed for up to 8x smaller vectors*Compressed from the first write
Savings stay yoursBuilt so the compression savings go to you
Built so the compression savings go to you.
Why now
Almost everyone uses AI. Many can't afford the data bill.
89% of respondents say their organisations regularly use AI in at least one business function, and 44% say AI is scaling across their enterprise (McKinsey, The State of AI, 2026). 42% of companies said they had abandoned most of their AI initiatives before production (S&P Global Market Intelligence, 451 Research, 2025). The retrieval backend decides both the bill and the accuracy.
02
"payment failed after retry"
ERR-4412clause 14.2SKU-88-A
who signed whatwhat depends on what
Meaning alone is not enough
Vector search finds similar text. Real work also needs relationships: clauses, owners, dependencies.
03
uncompressedCEREBRAL
Storage cost compounds
Embeddings grow with every document. CEREBRAL is built to compress from the first write, so more vectors can fit on less hardware.
Why CEREBRAL
Vector + graph, wherever it has to run.
CEREBRAL puts vectors and the graph in one engine, built to run in the cloud or on your own servers, including sovereign deployments.
Use cases
What builders ship on CEREBRAL. Five starting points.
Plain jobs for RAG and AI search. Regulated, on-prem and board-level needs live on the Enterprise page.
01 Domain search
Answers from your specialist information
Search over the corpus your experts already trust.
Policies, playbooks, research notes, clinical guidelines, legal opinions. CEREBRAL indexes the material that makes your organisation distinct and serves it back through search and assistants.
- Grounded in documents your team maintains
- Built for precise retrieval
example
clinical-guidelinesexamplesemantic
What does our protocol say about stage-two escalation?
Escalate to the on-call lead within the same shift once two criteria are met1, and record the handover in the case notes2.
protocol.md §4.2case-notes/handoverowner: clinical ops
02 Smart search
Ask in plain words. Find the right page.
Internal docs, policies and wikis, searchable like a conversation.
Nobody should have to know which folder held last year's policy. CEREBRAL is built to turn everyday questions into ranked answers across the content your organisation already has.
- Natural-language questions, ranked results
- Built for daily use
example
search · all sources4 results
how do I get travel for large trips approved?
PDFbest match
Travel & expenses policy
policies/travel-2025 · §3.1 approvals
WIKIwiki
Finance approvals matrix
wiki/finance/approvals
DOCform
Request form: travel exceptions
forms/travel-exception
MDfaq
FAQ: booking and approvals
help/travel-faq
03 RAG for AI assistants
Copilots that answer from real company data
Store your document embeddings, query by meaning, and pull connected records through the graph when you need them.
Retrieval-augmented generation only works when retrieval is trustworthy. CEREBRAL is built to give chatbots, copilots and agents the passages they need, so responses can cite what your organisation wrote. Agents connect through the MCP server.
- Built to return the passages your assistant needs
- The model talks; your data decides
assistant · retrieval onrag
Can contractors access the staging environment?
Yes, with a time-limited account approved by the system owner1. Production access stays restricted to employees2.
[1] access-policy §2.4[2] security-standard §6
grounded in 2 retrieved passages
04 Connected search
Meaning, then relationships
Find the passage by meaning, then see what it connects to.
Vector search finds what the text means. Relationships show how tickets, documents and accounts connect. CEREBRAL is built to find passages by meaning, then follow the relationships between them, in one engine.
- Meaning first, then graph hops
- Vectors and relationships in one engine
q: payment failed after retry · example
vectorcos
ticket #8812
retry guide
billing FAQ
timeouts
graphhops
#8812 → raised_by → Acme
Acme → account → plan
#8812 → related → #8790
example data
05 Cost at scale
Compressed from the first write
Designed for up to 8x smaller vector payload.* Built to fit more data on less hardware.
CEREBRAL compresses data the moment it's written, indexes and searches it while compressed, and is built to serve it from disk rather than RAM, which is where much of the hardware cost sits.
- No separate compression job
- Built so the compression savings go to you
designed for up to8x
designed for up to 8x smaller vector payload, from the first write
See it work
Five scenes, one engine. Watch a query move through CEREBRAL.
How it works
One API. Vector + graph. Compressed write path.
01
Compress on write
Vectors are compressed the moment they're stored, in the storage layer.
02
Built to search while compressed, from disk
Indexing and similarity search are built to run on the compressed data directly, served from disk instead of RAM. Built to search without unpacking per query.
03
Meaning, then relationships
Vectors and relationships in one engine, through one API: find by meaning, then follow relationships where you need them.
Illustrative · SDK coming soon
// Illustrative only. SDK coming soon; not yet published.// Today: self-hosted server (REST, gRPC, MCP) or the crbrl-pg Postgres extension. const db = cerebral.connect("http://localhost:8080"); // your server const result = await db.search({ collection: "contracts", query: "who owns clause 14.2 for Supplier A?", mode: "graph", // "semantic" | "graph" graph: { follow: ["has_clause", "owned_by"], depth: 2 }, topK: 8, include: ["text", "edges"],}); result.hits[0].edges // → example path# Illustrative only. SDK coming soon; not yet published.# Today: self-hosted server (REST, gRPC, MCP) or the crbrl-pg Postgres extension. db = cerebral.connect("http://localhost:8080") # your server result = db.search( collection="contracts", query="who owns clause 14.2 for Supplier A?", mode="graph", # "semantic" | "graph" graph={"follow": ["has_clause", "owned_by"], "depth": 2}, top_k=8, include=["text", "edges"],) result.hits[0].edges # → example path# Illustrative request shape. Your self-hosted server; no hosted API yet.curl "$CRBRL_URL/search" \ -H "Content-Type: application/json" \ -d '{ "collection": "contracts", "query": "who owns clause 14.2 for Supplier A?", "mode": "graph", "graph": { "follow": ["has_clause", "owned_by"], "depth": 2 }, "top_k": 8, "include": ["text", "edges"] }'example 200 · hits[0] contracts/2291 §14.2 → owned_by → Legal · sample output
Deploy anywhere
Your cloud. Your servers. Same CEREBRAL.
Run it where your app already lives. On-prem and sovereign options are on the Enterprise page.
region aCloud
primaryCEREBRAL
region bReplica
Cloud
Run CEREBRAL in the environment your team already operates. Vectors and relationships in one engine.
your data centre
On-prem & sovereign
On an enterprise licence: built for on-prem and sovereign deployment, including in Australia, for law, finance, health and government. Compressed indexes keep hardware lean built so the corpus can stay inside your boundary.
Talk to us → Mission
Lower the storage and cost barriers to AI search, so more people and local economies can innovate.
Serious retrieval infrastructure should not only be affordable for the largest labs. By compressing from the first write, CEREBRAL puts fast, accurate AI search within reach of teams on tighter budgets, including those keeping data on their own soil.
a Precog Labs product
Start with a pilot
Prove it on your own data. Then decide.
Pricing is set per deployment and confirmed on a call. Tell us about your data and where it must live, and we'll scope a pilot with you.
Need on-prem, regulated industries or a perpetual licence? Talk to Enterprise →
Early access
Request early access. Scale when it works.
Self-serve opens soon. Request early access and you're in the first wave, or request a pilot on your own data.