API
An OpenAI-compatible API for living business knowledge.
Call EigenVertex like a language model. Get answers grounded in your corpus, graph, memory, citations and business rules.
Why not just call a generic LLM?
A generalist model doesn’t know your sources, your decisions, your meetings, your constraints or your evidence. Every call starts from zero. EigenVertex adds the business-intelligence layer underneath the same call shape.
EigenVertex is not
- A thin wrapper around someone else’s model.
- A chatbot embedded in a widget.
- A vector database with a search box on top.
- A RAG demo that forgets context between calls.
- A fine-tune you have to retrain every time the business changes.
EigenVertex is
- A living domain-intelligence engine, callable like a model.
- A sovereign AI application layer your teams actually control.
- A surface familiar to anyone who has called an OpenAI-compatible API.
- A way to ship serious business applications without training your own model.
- A more governable, more confidential alternative to a bare LLM call.
OpenAI-compatible, EigenVertex-powered
Call EigenVertex exactly like you’d call an OpenAI-compatible endpoint. The shape is familiar. What comes back isn’t.
from openai import OpenAI
client = OpenAI(
base_url="https://api.eigenvertex.com/v1",
api_key="EVTX_API_KEY"
)
response = client.chat.completions.create(
model="eigenvertex/domain-intelligence",
messages=[
{"role": "user", "content": "What changed in this project since last briefing?"}
],
extra_body={
"workspace": "strategic-corpus",
"scope": "project",
"citations": True,
"graph": True
}
)A serious example
Same call shape, run the way it would actually run in production: scoped to a workspace, routed through retrieval and reasoning controls, and shaped into a decision briefing rather than a paragraph of prose.
Request
from openai import OpenAI
evtx = OpenAI(
base_url="https://api.eigenvertex.com/v1",
api_key=os.environ["EVTX_API_KEY"],
)
response = evtx.chat.completions.create(
model="eigenvertex/domain-intelligence",
messages=[
{
"role": "system",
"content": (
"You are preparing a decision briefing for the executive team. "
"Use citations, project memory, graph relationships and open questions."
),
},
{
"role": "user",
"content": (
"What changed since the last briefing, what decisions are now blocked, "
"and what should we review in tomorrow’s meeting?"
),
},
],
extra_body={
"workspace": "piano-heritage",
"scope": {
"corpus": ["technical-documents", "meeting-notes"],
"project": "supplier-transition-2026",
"session": "weekly-steering",
},
"retrieval": {
"vector": True,
"graph": True,
"temporal_memory": True,
"parent_expansion": True,
"rerank": {
"provider": "jina",
"model": "jina-reranker-v2-base-multilingual",
"top_k": 8,
},
},
"reasoning": {
"canonical_query": True,
"multi_query": True,
"self_reflective_evidence_score": True,
"minimum_evidence_score": 3,
},
"output": {
"format": "decision_briefing",
"include_citations": True,
"include_open_questions": True,
"include_action_items": True,
"include_graph_paths": True,
},
},
)
print(response.choices[0].message.content)Response
{
"answer": "Three supplier risks changed since the last briefing...",
"evidence_score": 4,
"citations": [
{
"source": "Supplier review - June 2026.pdf",
"page": 12,
"chunk_id": "chk_8f1...",
"quote": "Lead times increased from 6 to 11 weeks..."
}
],
"graph_paths": [
{
"from": "Supplier A",
"relation": "blocks",
"to": "September launch milestone"
}
],
"open_questions": [
"Has finance approved the alternative supplier budget?",
"Do we have updated delivery guarantees?"
],
"action_items": [
{
"owner": "Procurement",
"task": "Request updated lead time confirmation",
"due": "2026-06-28"
}
]
}You call it like a model. What comes back is a governed, sourced, graph-aware business deliverable — an answer, an evidence score, citations down to the page and the quote, the graph paths behind it, the open questions it surfaces, and the action items it implies.
What the API can use
Capabilities
Security, control and deployment
EigenVertex can sit as a governed layer between your applications, your data and the models you choose.
Build without fine-tuning
Fine-tuning has its place — but it is often rigid, costly and hard to govern. EigenVertex specialises behaviour a different way: through corpora, graphs, temporal memory, retrieval strategy, agent orchestration, query rewriting, reranking, structured outputs, rules and scopes.
Use cases