Daedalus¶
Kubernetes-native agent orchestration platform - dispatches work to ephemeral AI agent workers via message queues, using the A2A protocol for structured communication and KEDA for elastic scaling.
flowchart LR
User([User · Mattermost]) -->|A2A tasks| Orch[Orchestrator]
Orch -->|A2A envelope| NATS[(NATS JetStream)]
NATS --> Worker
subgraph Worker[Worker Pod]
direction TB
subgraph Platform["PLATFORM LAYER · Daedalus"]
Proxy["Queue-to-ACP Proxy sidecar<br/>reads queue (A2A) · writes results back"]
end
subgraph UserLayer["USER LAYER · bring your own"]
direction TB
Agent["ACP-compatible agent<br/>copilot --acp (default)<br/>claude --acp · codex --acp<br/>gemini --acp · qwen --acp<br/>+ 12 more ACP agents"]
Wrapper["Optional Layer 2 wrappers<br/>acpx (sessions, flows)<br/>copilot-bridge (hooks, personas)"]
end
Proxy -->|ACP| Agent
Agent -.->|optional| Wrapper
end
Proxy -->|results| NATS
classDef platform fill:#e8f0ff,stroke:#3b6fd1,color:#000
classDef user fill:#f3f0ff,stroke:#7a5cd1,color:#000
class Platform,Proxy platform
class UserLayer,Agent,Wrapper user
What is Daedalus?¶
Daedalus is the platform layer for running AI agents as Kubernetes-native workloads. It provides queue-based task dispatch, KEDA-driven autoscaling, agent discovery, and observability - and stays runtime-agnostic. Bring your own agent: any ACP-compatible CLI (Copilot, Claude, Codex, Gemini, Qwen, and a dozen more) drops in unmodified. The platform handles the boring parts (queueing, scaling, tracing, retries) so the agent layer can focus on the work.
Explore¶
-
:material-rocket-launch: Getting Started
Stand up Daedalus on AKS in under an hour.
-
:material-sitemap: Architecture
The four-layer model, ACP vs A2A, and the runtime contract.
-
:material-tools: Operations
Runbooks, observability, alerts, SIGTERM behavior.
-
:fontawesome-brands-github: Source
Issues, PRs, and roadmap on GitHub.