Overview
Kilimanjaro is a modular developer platform for building secure enterprise applications. You define AI agents, functions and data pipelines on the platform. The SDKs let you deploy and invoke them, read and write your data, publish to and subscribe to streams, sign users in, and encrypt with keys you hold.
Architecture at a glance
Kilimanjaro can be deployed in supported regional public-cloud, private-cloud or air-gapped environments. Your application reaches it through the SDK or the k9o CLI. Both connect to the API Gateway, which exposes Keep, Pulse, Gate, Kibo and the Governed Database. Data changes, egress requests, and tool calls are recorded in the audit trail, including approvals and blocks.
Functions and agents configured for platform-hosted execution run in Firelock, the open-source sandbox maintained by Kilimanjaro. Each Firelock execution runs in a Firecracker microVM with no network interface, and all guest egress is routed through one auditable vsock channel, with Keep's Warden layer evaluating each call. External agents can also submit tool calls through the platform's MCP interface, where Warden applies the same governance framework.
In BYOK and HYOK modes, your key remains in your own KMS or HSM. With the Kibo client core configured, data you encrypt with Kibo is encrypted and decrypted in your application environment; the platform only ever receives it as ciphertext.
Building blocks
| Block | What it does | Read |
|---|---|---|
| Keep | AI security and governance. Guardrails and policy for the agents you run. | Keep |
| Pulse | Data intelligence and real-time stream processing. Streams, analytics and live feeds. | Pulse |
| Gate | Customer identity and access management. Sign-in and sessions, and automatic data scoping so a signed-in user reaches only their own data. | Gate |
| Kibo | Cryptographic keys and secrets management. Client-side encryption with keys you hold, and a secrets vault. | Kibo |
| Governed Database | The operational data engine. Audit, data residency, access and provable erasure are enforced by the engine, not assembled in application code. | Governed Database |
Keep, Gate and Kibo each work on their own, in front of the backend you already have, and compose when you adopt more than one. All of them share one API contract, one identity and organisation model, and one audit trail.
How a request flows
An authenticated API call passes through the same layers whatever it is for.
Every layer is mandatory. There is no all-bypass key, no admin override and no service mode that skips role and policy checks. Platform-internal calls authenticate through Gate and run the same checks. See Unified Policy and the Identity-Aware Kernel.
Where your data lives
- Regions. A project's data is pinned to a region you choose when you create the project. The control plane, which holds organisation, project and member metadata, billing and the audit log, sits in a region you choose when you create the organisation. See Regions & residency.
- Keys. Every project has a custody mode that decides where its encryption keys live: Platform-Managed, SKR, BYOK or HYOK. In BYOK and HYOK the platform never holds your plaintext key, and code written for one mode runs unchanged against another. See Custody architecture.
Go deeper
- Sovereign Stack: the idea behind the platform and where it sits relative to sovereign cloud.
- Data architecture: the project as the isolation unit, and one query surface.
- Extensions: how an extension is deployed, and what it can reach.
- Agentic Compute: how agents and functions run.
- Security Model: the threat-model surfaces and the cryptographic foundation.
- System Specifications: slugs and routing, sessions, regions, the error envelope, billing and the error catalogue.
- The Technical Blueprint is the engineering design document, for architectural reviews and regulated workloads.