One graph. Every version. Every memory.

Kumiho is a graph-native system for versioning creative assets and persistent AI memory. It tracks what changed, when, why, and what depends on it without moving your files.

What is Kumiho? A versioned knowledge graph. For creative teams, it is version control for assets. For AI builders, it is persistent cognitive memory. Same graph, same API, same provenance guarantees.

Technical: Neo4j graph + Redis working memory. Raw artifacts remain on your own storage.

Get Started Free Watch Demo

Creative Asset View

Revisions, lineage, and metadata.

AI Memory Graph View

kref://CognitiveMemory/preferences/ui-theme.preference
r1 dark mode
r2 Dracula theme
r3 SUPERSEDES r2 -> Solarized

What you get

The core primitives that power everything.

Immutable Revisions

Every output becomes a revision. Published revisions are immutable; change happens via new revisions.

Typed Dependencies

DEPENDS_ON, DERIVED_FROM, SUPERSEDES and related edges encode why objects are connected.

Hybrid Retrieval

Full-text, graph traversal, and vector similarity fuse into one ranked retrieval model.

Automation & Events

Every change emits events. Trigger downstream jobs, workflows, and agent actions automatically.

Dream State

Async consolidation enriches and links memory with safety guards and bulk-operation circuit breakers.

How it works

Your data stays put. Kumiho tracks the rest.

Step 1

Your Storage

Local / NAS / S3 / Agent local disk

Files and conversation artifacts stay here

Step 2

Clients

MCP / Plugins / SDK / REST API

Step 3

Kumiho Graph

Neo4j + Redis

Relationships, metadata, working memory

Step 4

Query & Act

UI / Search / Events / Recall

Browse, trace, automate, remember

For Creative Teams

Version control without moving your files

Before -> After

Before: project_final/hero_v3_approved_v2.png
After:  kref://project/characters/hero.image
5 revisions · lineage · metadata · publish state

Real workflow example

Generate 200 AI variations with prompt/model/seed capture.
Compare and filter by metadata and lineage.
Pick and publish immutable revisions.
Trace provenance at any time.

See it in 60 seconds

For AI Builders

Persistent, auditable memory for AI agents

Kumiho gives agents durable memory that survives session resets and model switches while preserving provenance.

Step 1

Ingest

Step 2

Recall

Step 3

Revise

Step 4

Consolidate

Before: stateless agents
Session 1: "I prefer dark mode"
Session 2: asks again
Session 5: wrong preference recalled

After: persistent memory graph
kref://CognitiveMemory/preferences/ui-theme.preference
r1 dark mode
r2 Dracula theme
r3 SUPERSEDES r2 -> Solarized

One architecture. Shared primitives.

kref:// Addressing

Creative: Address any asset revision by URI

AI Memory: Address any memory revision by URI

Immutable Revisions

Creative: Published renders stay immutable

AI Memory: Beliefs evolve through SUPERSEDES revisions

Typed Edges

Creative: Know dependency breakage upstream/downstream

AI Memory: Trace evidence and reasoning lineage

BYO Storage

Creative: PSD/FBX/EXR remain on your storage

AI Memory: Full conversations remain local as artifacts

Hybrid Retrieval

Creative: Search by metadata + relation + semantics

AI Memory: Recall by keywords + graph path + semantics

Events & Automation

Creative: Re-render pipelines on source updates

AI Memory: Consolidate memory on session end

Patent pending. Same graph engine and provenance guarantees for both tracks.

Ways to start

Pick your entry point based on how you work.

Create

ComfyUI

Version every generation. Capture prompt, model, seed, and lineage.

Set up ComfyUI nodes

Automate

n8n

Trigger downstream work when assets or memory states change.

Explore n8n workflows

Connect

MCP

Give MCP-compatible agents persistent memory with a single integration path.

Read MCP integration guide

Build

Python SDK

Embed revision history and memory primitives into your own product.

Read SDK docs

Integrate

REST API

Direct HTTP access to items, revisions, edges, and search pipelines.

Browse API reference

Built for scale, designed for trust

Multi-tenant isolation

Data residency

Real-time events

Safety guards (Dream State circuit breaker)

Patent-pending architecture

Immutable audit trail

Common questions

Do I have to migrate storage?

No. Files stay where they are.

How is this different from a vector DB?

Kumiho adds typed revisions, provenance, and dependency edges.

Does conversation data leave my machine?

Full conversations can remain local as artifacts.

Can I switch LLM providers?

Yes. Memory is model-decoupled.

Is Kumiho open source?

kumiho-python and MCP tooling are open source.

What does free tier include?

Core graph workspace with AI Cognitive Memory included.

Ready to give your work and your AI a memory?

Join creators, studios, and AI builders who trust Kumiho to track what matters.

Get Started Free Read the Docs

Free tier available. Start with one folder or one MCP connection.