Introduction
Klyro is a composable stack generator for self-hosted AI agents. You choose a runtime and integrations, Klyro validates compatibility, and you get a reproducible Docker Compose deployment you own. It is a CLI-first, declarative compiler: not a runtime and not a control plane.
The problem Klyro solves
Section titled “The problem Klyro solves”Self-hosted agent runtimes are increasingly composable, but their integrations, configuration formats, persistence requirements, and deployment conventions stay fragmented. Expressing a simple intent such as “run Hermes with Mnemosyne memory, Telegram, a WebUI, and OpenRouter” still requires knowing which integrations are native, which plugins connect them, which environment variables and volumes must exist, and how to express all of it as deployable infrastructure.
Klyro turns that intent into a validated, deterministic deployment plan and deployable artifacts:
klyro.yaml -> decode + resolve -> plan -> compose project you ownWhat Klyro is not
Section titled “What Klyro is not”Klyro is not an agent runtime, scheduler, daemon, container control plane, model gateway, secret manager, or hosted service required by your stack. If Klyro had to remain running after your stack starts, that feature would be out of scope. The full boundary is documented under Limitations.
v0.1 scope at a glance
Section titled “v0.1 scope at a glance”| Area | Supported in v0.1 |
|---|---|
| Runtime | Hermes |
| Memory | Hermes built-in, Mnemosyne |
| Interfaces | Telegram, Hermes WebUI |
| Model configuration | OpenRouter |
| Renderer | Docker Compose |
The scope is deliberate: every advertised combination ships with contract tests, reviewed golden output, and documentation. See Providers for the support matrix.
Where to go next
Section titled “Where to go next”- Installation: requirements and setup.
- Quickstart: generate your first stack.
- How it works: the decode, resolve, plan, render pipeline.