Agent memory, governed.
Kumiho gives your agents persistent memory across sessions, auditable action trails, and experience that compounds — on a versioned provenance graph your security and compliance teams control. Your cloud, your region, or fully on-prem.
dedicated Kubernetes · BYO-LLM & embeddings · SSO/SAML · on-prem option
Agents without memory can't be trusted to work
An agent that forgets everything between sessions isn't a worker — it's a calculator. Real agentic productivity requires memory, auditability, and accumulated experience.
- cold startAgents start blind every run
- No memory of past instructions, decisions, or failures. Every session reinvents the wheel — losing the context that made previous runs successful.
- unverifiableNo audit trail
- You cannot verify what an agent did, decided, or why. When something goes wrong — or right — there is no record to learn from or defend against.
- no compoundingNo accumulated experience
- Agents repeat mistakes instead of building on what worked. Without persistent memory, there is no way for the system to improve over time.
Infrastructure for every AI workflow you need to build
- Graph-native SDK & API
- Python, Dart, and C++ SDKs built on a rich graph schema, plus a stateless REST API for any client. Spaces, bundles, items, revisions — queryable, traversable, and extensible. Build tools that think in relationships.
- Asset tracking & full lineage
- Every asset is a revisioned node. Every parent is recorded. Track anything from first draft to final delivery — provenance is a first-class property, not an afterthought.
- Autonomous pipeline engine
- Connect asset and memory events to webhooks, agents, and downstream workflows. Pipelines that drive themselves — no custom glue code, no brittle triggers.
- Team governance & audit
- Role-gated access, SSO/SAML (Okta, Azure AD), full mutation history, and a complete record of every change. SOC2 compliance in progress.
Not a bolt-on memory tool
Most tools add memory or storage as an afterthought. Kumiho is a graph-native infrastructure platform built from the ground up for pipelines, lineage, and scale.
| Aspect | Others | Kumiho |
|---|---|---|
| Architecture | Vector search — flat, similarity-based retrieval | Graph-native — relationship-aware, traversable at any depth |
| SDK coverage | REST endpoints only, no typed SDK | Python, Dart & C++ SDKs + REST API — full graph query on every client |
| Pipeline automation | Manual triggers — custom glue code required | Event-driven, autonomous — pipelines drive themselves |
| Asset lineage | Not built in — retrofitted if at all | First-class — every node is a revision with provenance |
| LLM + infra flexibility | Locked to one provider or deployment model | BYO-LLM, BYO embedding, any cloud or on-prem |
What teams build on it
- Autonomous asset pipelines
- Trigger downstream agents and workflows automatically when assets are created, revised, or approved. No glue code required.
- AI agent memory systems
- Give any agent durable, queryable memory across sessions. Shared context across your entire team — graph-native from the ground up.
- Creative production tracking
- Track every asset from brief to final delivery with full revision history, prompt lineage, and approval chains.
- Compliance audit infrastructure
- Immutable records, role-gated access, and a complete mutation log. Prove exactly what happened, when, and who approved it.
- Multi-agent orchestration
- Coordinate fleets of specialised agents using shared memory and event-driven triggers. Pipelines that route themselves.
- Enterprise knowledge graphs
- Connect documents, decisions, artifacts, and people into a living graph. Query relationships the way your business actually thinks.
Deploy on your terms
Agent memory is governed data. Where it lives — and what leaves your perimeter — is an explicit, verifiable choice at every tier.
Where agent memory lives
- Self-hosted Community Edition
- The server binds to loopback only and stores everything in a local Neo4j. Full conversations can stay on the machine as markdown artifacts — nothing leaves it.
- Kumiho Cloud
- If you opt into cloud, only short structured summaries sync. Raw conversations and working files remain where you keep them.
- Enterprise
- Dedicated kumiho-server and a dedicated Neo4j instance in your cloud and region — or on your own Kubernetes cluster, fully on-prem, with one license key.
- Any cloud, any region
- Deploy on AWS, GCP, or Azure — in any region you choose. Dedicated kumiho-server on Kubernetes and a dedicated Neo4j instance. No shared infrastructure, no noisy neighbours.
- No model training on your data
- Your graph is your graph. We never use your data to train models or improve our systems.
- On-premises deployment
- Run kumiho-server on your own Kubernetes cluster with a dedicated Neo4j Enterprise instance and Redis on-prem. No shared infrastructure, no other tenants.
Controls & guarantees
- Dedicated kumiho-server on Kubernetes
- Dedicated Neo4j instance — no shared tenants
- BYO embedding model
- SSO / SAML — Okta, Azure AD
- 99.9% uptime SLA
- GDPR compliant
- JWT-gated per-tenant access
- Encryption at rest and in transit
- SOC2 Type II in progress
- Role-based access control
- On-prem Kubernetes deployment
Ready to give your agents a memory they can work from?
Let's talk about what productive agentic AI looks like for your team — persistent memory, auditable pipelines, and a system that gets better the more it works.
Or email us at support@kumiho.io
Custom pricing · Annual contracts · Volume licensing