About Evokoa

Petabyte-scale data, traversable by agents.

Enterprise systems generate more data than any team can read. Evokoa builds a relationship layer over that data so AI agents can traverse it in milliseconds. No pipelines. No copies. No second database.

EVOKOA
InputPostgres Schema
IndexRelationship Cache
RuntimeRust Hot Path
OutputFull Context

Where Evokoa Works

Six problems, one layer.

Every use case below has the same root cause: the relationships exist in the database, but nothing treats them as queryable objects.

01AI InfrastructureAgentic memoryTeams building agents that need to remember and reason
Product question

What does this agent already know about this entity?

The bottleneck

Agents forget between sessions. Vector stores retrieve semantically similar text but lose the structural relationships between customers, orders, conversations, and actions that define real context.

What Evokoa does

Evokoa gives agents a persistent, traversable memory of how every entity in the system connects, so they can recall context instantly without re-ingesting.

02AI ApplicationsAgentic product workflowsFounders and developers building agents on Postgres
Product question

What should this agent know before it acts?

The bottleneck

Making enterprise data available to AI means months of pipeline work, a second database, and an ETL team. Most companies never start. The ones that do spend more time maintaining the pipes than building the product.

What Evokoa does

Evokoa makes the data traversable where it already lives. The agent sees the full chain of relationships before it acts, with no data movement.

03Financial ServicesReal-time fraud detectionFintech teams, marketplaces, payment products
Product question

Is this transaction connected to a risky pattern?

The bottleneck

Fraud rings hide across many hops: devices, cards, accounts, addresses, merchants, and prior events. Recursive joins collapse at checkout speed.

What Evokoa does

Evokoa traverses the relationship graph while the transaction is still live. Products can block, step up, or approve with structural context, not just row-level rules.

04Internal ToolsSupport and operations copilotsTeams building support, success, and operations tools
Product question

What context explains this customer issue?

The bottleneck

Customers connect to tickets. Tickets connect to invoices. Invoices connect to approvals. The copilot needs to traverse this chain in milliseconds. Flattening it into a prompt is not enough.

What Evokoa does

Evokoa returns the connected context behind the customer so copilots can explain what happened and surface the right action.

05SecurityPermission-aware applicationsSaaS teams with complex account models
Product question

Can this user see or change this record?

The bottleneck

Permissions are relationships across teams, roles, resources, customers, workspaces, regions, and delegated access. No single column captures this.

What Evokoa does

Evokoa resolves multi-hop access paths quickly while Postgres remains the authority for the underlying records.

06InfrastructureDependency and blast radiusDeveloper platforms, logistics systems, infra teams
Product question

What breaks if this node changes?

The bottleneck

Modern systems depend on chains of services, suppliers, records, owners, and workflows. Batch reports are stale the moment they are generated.

What Evokoa does

Evokoa provides a live relationship map for impact analysis, dependency checks, and automated recovery workflows.

The Relationship Layer

Not another database. A layer that makes your data traversable.

The layer that sits between your existing systems and the AI that needs to reason over them. A single API that makes your entire company traversable. By humans, by code, by agents. No data movement. No second source of truth.

Source systems

Rows stay in Postgres

Evokoa

Maps IDs and relationships
Virtual graph

App or agent

Asks about connected data

Matching IDs

Only relevant rows are fetched

Trusted source data returns to the app

The Architecture

Topology in memory. Data at rest.

Topology

The graph lives in memory. The data stays in Postgres.

Evokoa builds a compact in-memory index of how your records connect. It stores IDs and edge types, not row payloads. When a traversal finds matching paths, Evokoa hydrates the relevant rows from your existing database.

Scale

~34x less RAM than traditional graph databases on the Panama Papers dataset.

By separating topology from data, the relationship index stays small enough to fit in memory even at petabyte-scale workloads. The source database does what it already does well: store and serve rows.

Runtime

Rust on the critical path. No GC pauses.

The traversal engine is written in Rust. No garbage collector. No JVM. Agent workflows are latency-sensitive, and a relationship lookup that stalls for GC is a relationship lookup that ships too late.

Evokoa vs Apache AGE

Different layer, different job.

Apache AGE and Evokoa both bring graph-style querying to Postgres, but the architecture is different. AGE converts Cypher into recursive calls inside Postgres, which can break down as paths get deeper. Evokoa creates a virtual graph layer in memory, so Postgres remains the query interface while Evokoa handles deep relationship traversal, including 10+ hop paths.

Different execution model

Apache AGE translates Cypher into recursive work inside Postgres. That can degrade quickly as traversals get deeper. Evokoa builds a virtual graph layer in memory for relationship traversal.

Built for deep traversal

AGE-style recursive calls can start dying around multi-hop workloads. Evokoa is designed for deep traversals, including 10+ hop paths, because the hot path is served from the in-memory graph layer.

Postgres remains the interface

Evokoa is being open sourced as a Postgres extension, so builders can query inside Postgres while Evokoa maintains the virtual relationship graph needed for scale.

What We Believe

Principles, not slogans.

No data movement

Your database is the source of truth. It stays that way. Evokoa never copies row data into a second store.

Traversable by default

If the data exists in your system, agents should be able to walk the relationships immediately. Not after a migration.

Ship, then optimize

We build for developers who need working infrastructure this week, not a six-month data program.

Specificity over narrative

~34x less RAM. 10+ hop traversals. Microsecond lookups. We prefer numbers to adjectives.

Get Started

Every company will be AI-native within a decade.

Point Evokoa at your existing database. We build the relationship graph in memory. Your agents traverse it in milliseconds. Your data never moves.

Join UsRead the Manifesto →
NewWe have just launched pgGraph, our open source postgres extensionLearn more