Most AI projects stall at the data layer, unstructured documents, siloed databases, no shared schema. Perseus turns that into a knowledge graph your AI can actually reason over.
"We'd stalled on getting our knowledge graph production-ready for months. With Perseus we had a reliable ontology and a working graph pipeline in under a month."
Magnus Helander
Read the case study
Each capability works independently or together, so teams ship quickly and scale on a reliable graph foundation.
Auto-generate a custom ontology from your source documents, the blueprint for your graph. Describe your use case in plain language, or upload and extend an ontology you already have (OWL, SKOS, custom).
Learn More
Turn raw documents into a structured network of entities and relationships. Automated text-to-graph pipelines produce standardized, production-ready graphs built to power AI applications.
Learn More
Return entities, relationships, and traversal paths, not text chunks. Hybrid vector + graph retrieval so agents reason over connected structure, with explicit source attribution on every result.
Learn More
From Fortune 500s to fast-moving startups, customers are building the next generation of AI applications on structured knowledge.
Our fine-tuned model outperforms every frontier model on text-to-graph extraction — reliability, accuracy, and cost.
99.95%
Output reliability
89%
Average F1 score
+9.5 points
vs GPT 5.5
+37.8 points
vs Opus 4.8
Lettria combines benchmark-backed performance with external validation and real-world production deployments.
An ontology gives every silo one shared schema, so scattered documents and databases resolve to the same entities instead of a thousand duplicates. The graph makes the connections queryable; the ontology keeps them consistent as the domain grows.
See the Use case
Conversation history forgets; a graph remembers as structure. The ontology defines what your agent knows about the world, so it reasons over typed relationships that persist and update across sessions — not a scrolling transcript.
See the Use case
Chunk retrieval finds text near the answer. An ontology-powered graph retrieves the relationships that are the answer, each with a traceable path back to source — the difference between plausible and grounded.
See the Use case
Modeling a graph correctly is the hard part, the right entities, the right relationships, evolving them as the domain changes. When your team wants that expertise embedded, our Forward Deployed engineers and ontologists work alongside you. Optional, always, an accelerant, not a requirement.
Launch your first graph-powered flow quickly, then scale to enterprise throughput with the reliability, support, and infrastructure options production teams need.
Develop
(REST & Python SDK)
Execute
(Serverless compute)
Scale
(Enterprise ready)
Monitor
(Console & alerts)
Choose the plan that fits your first graph workflow, then scale with your usage.
No time limit
No credit card
Fair-use volume limits
Workload capacity + Committed usage
VPC deployment
DPA available
On-prem / air-gapped
Full security docs
SOC 2 Type II, GDPR (with DPA), and ISO 27001. HIPAA-ready configuration and EU AI Act-aligned controls are available on Enterprise and above. Full compliance documentation including pen-test summaries is available under NDA on the Security and trust page ↗.
Perseus is the graph infrastructure layer underneath Lettria. You do not interact with it directly on the Lettria platform, but its reliability is the reason Lettria's accuracy and explainability hold up in production. If your platform team also wants SDK-level access to build custom knowledge graphs, see Perseus ↗.
A typical engagement opens with a two-week ontology sprint, followed by graph build over 6 to 12 weeks, then ongoing refresh cycles. Our ontologists work embedded with your subject-matter experts, not at arm's length, and transfer knowledge as they go so you are not locked in.
Yes, this is how most of our customers start. Pilot terms convert into Enterprise agreements with pilot fees credited against the first year's license. No rebuild of your ontology or workspace is required.
Annual invoicing against a PO by default. Multi-year agreements and custom billing cycles are supported on Enterprise. On the + Services tier, platform license and services hours are billed as separate line items so services procurement can approve them independently.
Documents and derived graph data are retained for the duration of your contract and deleted within 30 days on request or on termination. Audit-log retention is configurable on Enterprise to match your internal compliance window.
VPC is available on Enterprise. On-prem and air-gapped deployments are available on Enterprise + Expert Services, including for sovereign-cloud requirements. Our team handles installation and lifecycle, you do not need in-house graph-database expertise.
Data is stored in the region you select (EU, US, or custom for Enterprise) and never leaves it. Your documents, your ontology, and your queries are never used to train our models or any third-party model. Our DPA contains the full commitment in writing.
Choose the path that fits your team: explore Knowledge Studio or start building with Perseus.
.jpg)