Aivarium
Aivarium
Vector search finds what sounds similar. Aivarium finds what's actually connected. Graph-native ingestion and multi-hop traversal for AI systems that need to reason over complex, relationship-dense data.
import aivarium
# Ingest any data source
client = aivarium.Client(api_key="aiv_...")
client.ingest(source="./legal-docs", type="pdf")
client.ingest(source=conversation_history, type="str")
# Multi-hop graph retrieval
result = client.query("Which clauses govern liability in Schedule B?")
print(result.graph)
# {
# "nodes": [
# {"id": "n1", "label": "Contract A", "type": "Document", "source": "contract_a.pdf"},
# {"id": "n2", "label": "Schedule B", "type": "Section", "source": "contract_a.pdf"},
# {"id": "n3", "label": "Clause 7", "type": "Clause", "source": "contract_a.pdf"},
# {"id": "n4", "label": "Liability Cap", "type": "Concept", "source": "contract_a.pdf"},
# {"id": "n5", "label": "Regulation §4.2", "type": "Regulation", "source": "reg_2024.pdf"}
# ],
# "edges": [
# {"from": "n1", "to": "n2", "relation": "contains"},
# {"from": "n2", "to": "n3", "relation": "includes"},
# {"from": "n3", "to": "n4", "relation": "defines"},
# {"from": "n4", "to": "n5", "relation": "governed_by"}
# ],
# "hops": 4
# }
print(result.context)
# "Schedule B (Section 3.2) contains Clause 7, which defines the Liability Cap
# as a maximum of £2M per incident. This cap is subject to Regulation §4.2
# (2024 Financial Services Act), which mandates a minimum coverage floor of
# £500K. Clause 7 takes precedence over Schedule B general terms in the
# event of a conflict, per Contract A Section 1.4."Legal documents, medical records, codebases, and complex conversations aren't just bags of text. The structure between concepts is where the intelligence lives.
When you query Aivarium, an agent traverses the graph across multiple hops — following relationship chains that a semantic similarity search would never find. Every interconnected concept is returned, not just the nearest neighbours.
Upload raw documents, PDFs, markdown, or conversation history. Aivarium uses LLM-powered entity and relationship extraction to build a structured knowledge graph automatically — no schema design required.
Every node and edge is tagged with its source, creation timestamp, and data origin. Incorrect data? Roll back to any prior state. For law, medicine, and compliance — auditability is a first-class feature.
Watch your knowledge graph grow in real time as data is ingested. See nodes appear, edges form, and clusters emerge. Full query traceability — see exactly which nodes were returned and why.
Python and JavaScript SDKs. REST API. VS Code extension. Plug directly into OpenAI, Anthropic, or any LLM. Designed to drop into your existing RAG pipeline in minutes.
Upload documents, PDFs, markdown files, or stream conversation history directly via SDK or the web interface. Aivarium accepts raw, unstructured data — no pre-processing required.
client.ingest(source="./contracts", type="pdf")
client.ingest(source=chat_log, type="str")An LLM extracts entities and relationships from your data, building a structured knowledge graph with typed nodes and edges. Every element is tagged with its source and timestamp. Watch it build live in the dashboard.
At query time, an agent traverses the graph — not just the nearest semantic neighbours, but the full chain of connected concepts. Returns structured graph context plus a summarised answer, with full traceability of every hop.
result = client.query("What regulations affect Party B's obligations?")
# Traverses: Party B → Contract A → Regulation §4.2 → Clause 7
# Returns all 4 connected nodes across 3 hopsCross-reference contracts, statutes, clauses, and precedents. Find how a regulatory change three hops away affects a specific obligation — something no similarity search can do.
Map relationships between conditions, treatments, contraindications, and patient history. Retrieve all clinically relevant context, not just the symptom-matching chunks.
Our VS Code extension replaces GitHub Copilot with a graph-powered assistant. As you code, Aivarium builds a live knowledge graph of your entire codebase in the background. Unlike Copilot, it never loses context as your chat grows — it traverses the graph, not a shrinking window.
Install the extension →Ingest conversation history as a graph of concepts, decisions, and commitments. Your AI agents remember everything — not just the last N tokens.
Designed to replace your vector DB or sit alongside your existing pipeline.