Knowledge Graph

AEGIS stores oil and gas entity relationships in an Apache AGE property graph running inside PostgreSQL 15. All graph queries use openCypher syntax.

Technology Stack

Component	Details
Graph engine	Apache AGE (A Graph Extension for PostgreSQL)
Query language	openCypher
Host database	PostgreSQL 15
Graph name	`oilgas`
Service	Knowledge Graph Service (port 8003)

Setup Requirements

Every connection to Apache AGE must execute two preparatory statements before running Cypher queries:


LOAD 'age';
SET search_path = ag_catalog, "$user", public;

The AgePool connection helper in knowledge_graph/age_connection.py handles this automatically for every acquired connection.

Graph Schema

Vertex Labels

The graph supports the following vertex types:

Label	Description	Key Properties
`Well`	Oil or gas well	`api_number`, `name`, `status`, `well_type`, `spud_date`, `surface_location`, production fields
`Lease`	Mineral lease	`entity_id`, `name`, `rrc_lease_id`, `county`, `district`
`Field`	Oil/gas field	`entity_id`, `name`, `rrc_field_id`, `district`, `county`, `state`
`Formation`	Geological formation	`entity_id`, `name`, `type`, `depth_range`
`Operator`	Operating company	`entity_id`, `name`, `rrc_operator_id`, `p5_number`, `contact_name`, `contact_email`
`Permit`	Drilling or operating permit	`entity_id`, `permit_number`, `status`, `type`
`FlaringAuthorization`	R-32 flaring authorization	`entity_id`, `auth_number`, `authorized_mcf`, `expiry_date`
`FlaringEvent`	Flaring/venting occurrence	`entity_id`, `disposition_code`, `volume_mcf`, `date`
`Wellpad`	Surface grouping of wells	`entity_id`, `name`, `wells_count`
`Facility`	Processing/storage facility	`entity_id`, `name`, `type` (tank_battery, compressor, separator, cpf)
`PipelineRoute`	Pipeline segment	`entity_id`, `name`, `diameter`, `capacity`
`InfrastructureProject`	Large infrastructure project	`entity_id`, `name`, `status`, `type`
`Regulation`	Regulatory rule	`entity_id`, `name`, `rule_number`, `source`
`WellEvent`	Event on a well	`entity_id`, `event_type`, `date`
`ComplianceDeadline`	Regulatory deadline	`entity_id`, `deadline_date`, `type`
`FilingPackage`	Assembled filing	`entity_id`, `filing_type`, `status`
`WellConversation`	Chat session about a well	`entity_id`, `conversation_id`
`Project`	General project	`entity_id`, `name`, `status`
`Equipment`	Field equipment	`entity_id`, `name`, `type`
`Pipeline`	Pipeline entity	`entity_id`, `name`

Edge Labels

Edges define relationships between vertices:

Edge Label	From	To	Semantics
`LOCATED_IN`	Well	Field, Lease	Well is located in a field or on a lease
`LOCATED_ON`	Well	Wellpad	Well is physically on a wellpad
`COMPLETED_IN`	Well	Formation	Well produces from a geological formation
`OPERATED_BY`	Well, Lease, Wellpad, Facility, PipelineRoute	Operator	Entity is operated by a company
`OFFSET_TO`	Well	Well	Wells are within regulatory spacing distance
`GOVERNED_BY`	Well, Permit	Regulation	Entity is subject to a regulation
`FLARES_AT`	FlaringEvent	Well, Lease	Flaring event occurs at a well or lease
`AUTHORIZED_UNDER`	FlaringEvent	FlaringAuthorization	Event is covered by an authorization
`PRODUCES_TO`	Well	Facility	Well’s production flows to a facility
`FEEDS_INTO`	Facility	PipelineRoute	Facility feeds into a pipeline
`CONNECTS_TO`	InfrastructureProject, PipelineRoute	Lease, InfrastructureProject	Infrastructure connects to entities
`HAS_EVENT`	Well	WellEvent	Well has an associated event
`HAS_DEADLINE`	Well	ComplianceDeadline	Well has a compliance deadline
`HAS_FILING`	Well	FilingPackage	Well has a filing package
`HAS_CONVERSATION`	Well	WellConversation	Well has a chat session
`PART_OF`	(any)	(any)	Generic hierarchical parent-child relationship

Sample Cypher Queries

Find a well and its operator


MATCH (w:Well {api_number: '42-329-12345'})-[:OPERATED_BY]->(op:Operator)
RETURN w, op

Find all offset wells for a given well


MATCH (w:Well {api_number: '42-329-12345'})-[:OFFSET_TO]->(offset:Well)
RETURN offset.api_number, offset.name, offset.status

Trace production flow from well to pipeline


MATCH (w:Well {api_number: '42-329-12345'})-[:PRODUCES_TO]->(f:Facility)-[:FEEDS_INTO]->(p:PipelineRoute)
RETURN w.name, f.name, f.type, p.name

Find all wells on a lease


MATCH (w:Well)-[:LOCATED_IN]->(l:Lease {entity_id: 'lease-mitchell-ranch'})
RETURN w.api_number, w.name, w.status

Impact propagation (multi-hop)


MATCH path = (start:Facility {entity_id: 'fac-tank-battery-a'})<-[:PRODUCES_TO]-(w:Well)
RETURN w.api_number, w.name, w.production_oil_bbls, w.production_gas_mcf

Context Assembly (Tier 3.5)

The ContextAssembler class in knowledge_graph/context.py gathers entity context for injection into the agent pipeline. Given a well API number and optional domain tags, it assembles:

The well and its properties
Its lease, field, and operator
Active flaring authorizations
Offset wells (via OFFSET_TO edges)
Applicable regulations (via GOVERNED_BY edges)
Connected infrastructure (wellpad, facilities, pipeline routes)

Domain Tag Filtering

Domain tags control which sections are included in the assembled context:

Tag	Sections Included
`spacing` / `rule_37`	well, lease, field, operator, offsets, regulations
`flaring`	well, lease, field, operator, flaring_auths, infrastructure, regulations, wellpad, facilities, pipeline_routes
`rule_32`	well, lease, field, operator, flaring_auths, infrastructure, regulations
`compliance`	well, lease, field, operator, offsets, flaring_auths, regulations, wellpad, facilities
`drilling`	well, lease, field, operator, regulations
`production`	well, lease, field, operator, wellpad, facilities
`impact`	All sections

If no domain tags are specified, all sections are included.

Managed Entity Context

For the managed entity system (entity type definitions), context assembly is available via a separate endpoint that works with any entity type, not just wells:


GET /context/assemble/managed/{entity_id}?entity_type_key=well&domain_tags=spacing

API Endpoints

Entity CRUD

Method	Path	Description
`POST /entities`	Create a vertex	Body: `{label, properties}`
`GET /entities/{label}/{id}`	Get a vertex	By label and entity_id
`GET /entities/{label}`	List vertices	By label (limit param)
`PUT /entities/{label}/{id}`	Update vertex properties	Body: `{properties}`
`DELETE /entities/{label}/{id}`	Delete vertex and edges

Edge CRUD

Method	Path	Description
`POST /edges`	Create an edge	Body: `{from_label, from_id, edge_label, to_label, to_id, properties}`
`GET /edges/{label}/{id}`	Get edges for a vertex	Query: `edge_label`, `direction` (out/in/both)

Raw Cypher

Method	Path	Description
`POST /query`	Execute raw openCypher	Body: `{query, columns}`

Context Assembly

Method	Path	Description
`GET /context/assemble/{well_api}`	Legacy well-centric context	Query: `domain_tags`
`GET /context/assemble/managed/{entity_id}`	Managed entity context	Query: `entity_type_key`, `domain_tags`

Impact Analysis

Method	Path	Description
`GET /impact/{label}/{entity_id}`	Bidirectional impact propagation	Query: `max_depth`, `wti_price`, `henry_hub_price`
`GET /impact/outward/{label}/{entity_id}`	Outward (downstream) impact
`GET /impact/inward/{label}/{entity_id}`	Inward (upstream) impact
`GET /impact/tree`	Full infrastructure tree

Seed

Method	Path	Description
`POST /seed`	Load sample Permian Basin data	Creates ~40 vertices and ~50 edges

Apache AGE stores all property values as strings internally. When reading properties back, you may need to parse numeric or JSON values from strings. The CRUD layer handles basic type coercion, but complex nested structures (like production_history_12m) are stored as JSON strings.