Last updated 08-14-2026
Category:
Reviews:
Join thousands of AI enthusiasts in the World of AI!
Lettria
Lettria builds a graph context layer that turns enterprise documents and databases into ontology-backed knowledge graphs for automation teams. Its Knowledge Studio GraphRAG product parses PDFs, tables, SAP exports, and scanned annexes, then answers multi-hop questions with source citations. Perseus is the developer platform for generating ontologies and graphs without hand-modeling every entity.
Vector RAG tools retrieve similar chunks; Lettria follows explicit relationships across structured and unstructured data. The company publishes benchmark comparisons showing graph agents at 81.7% overall accuracy versus 57.5% for vector agents, plus text-to-graph claims of 30% higher accuracy and up to 400x faster extraction than general LLMs. That trade-off favors regulated teams that need audit trails over quick semantic search.
Finance, healthcare, legal, and engineering groups use Lettria for ESG filings, biomedical literature, contract corpora, and technical manuals. Case studies cite Alfa Laval (+30% extraction accuracy), AP-HP (+60% faster research), and Wisecube processing 108 GB of biomedical text into 500+ monitored classes.
GraphRAG parses multimodal PDFs, tables, SAP exports, and scanned annexes into queryable graphs
Perseus Free tier includes 5 graph builds and 1 ontology build per 30 days on inputs up to 30 KB
Text-to-Graph benchmark claims 30%+ higher accuracy and up to 400x faster than general LLMs
Graph agents reach 81.7% overall accuracy versus 57.5% for vector agents on Lettria benchmarks
Enterprise deployments offer VPC, on-prem, air-gapped options with SSO, SAML, and audit logs
Pilot programs scope one use case in 8 to 12 weeks with embedded ontologists and graph engineers
GraphRAG targets tables, scans, and annexes that break generic document parsers.
Published benchmarks cite 81.7% graph agent accuracy with full answer provenance.
Perseus free tier lets developers prototype ontologies before enterprise GraphRAG sales cycles.
Deployment options span cloud, VPC, on-prem, and air-gapped environments.
Knowledge Studio GraphRAG requires scoped enterprise contracts, not self-serve checkout.
Perseus Free caps inputs at 30 KB and allows only five graph builds per month.
Complex regulated rollouts may need forward-deployed ontologists beyond platform software.
What does Lettria do?
Lettria unifies structured and unstructured enterprise data into ontology-powered knowledge graphs. Knowledge Studio GraphRAG answers complex document questions with traceable sources, while Perseus lets developers generate ontologies and graphs in minutes instead of months.
How much does Perseus cost on Lettria?
Lettria offers a free Perseus tier with 5 graph builds and 1 ontology build per 30 days on 30 KB inputs. The Plus plan charges $0.01 per PCU with 3,000 graph builds per month on 1 MB inputs. Enterprise Perseus is custom priced with unlimited builds.
How is Lettria GraphRAG priced?
Lettria sells GraphRAG through scoped enterprise engagements, not seat fees or per-query meters. Options include an 8 to 12 week pilot, production deployment across workspaces, and in-house expert services with embedded ontologists. Pricing is quoted per engagement.
Which industries does Lettria target?
Lettria focuses on finance, healthcare, legal, and engineering teams in regulated industries. Published case studies include Alfa Laval technical documentation, AP-HP biomedical research, Leroy Merlin product catalogs, and Wisecube biomedical knowledge graphs.
Does Lettria support on-prem deployment?
Yes. Lettria offers cloud, VPC, on-prem, and air-gapped deployments on enterprise tiers. Production and in-house expert plans list on-prem availability, and the FAQ notes sovereign-cloud requirements handled by Lettria engineers.
How does Lettria compare to vector RAG?
Lettria argues vector search breaks on multi-hop reasoning across fragmented chunks. Its published benchmarks show graph agents above 90% on multi-hop tasks and 81.7% overall accuracy, while vector agents score 57.5%, with full provenance on each answer.