Skip to content

Astromesh Leia

Astromesh Leia is a Claude Code plugin that gives you a natural-language interface — /leia ... slash commands — for creating, deploying, and managing AI agents on astromesh-nexus Kubernetes clusters. The goal is simple: go from a business idea to a deployed WhatsApp agent in minutes, with no Kubernetes expertise required.

Leia is named after a lemon beagle — approachable, loyal, and friendly. That personality carries into the plugin: you describe what you need in plain English, and Leia handles the cluster work behind the scenes.

  • A Claude Code plugin you install once and drive through /leia slash commands.
  • A conversational CLI: describe what you want in natural language, or use explicit subcommands.
  • An orchestrator that designs agent manifests, talks to the Nexus REST API over curl, and runs kubectl for cluster operations.
  • A batteries-included starter kit with business-vertical templates, specialized subagents, and bundled astromesh/v1 schemas.
  • It is not a SaaS — it runs locally inside your Claude Code CLI.
  • It is not a no-code GUI — it is a conversational command-line plugin.
  • It is not an agent framework — the agents it builds run on Nexus and the Astromesh runtime, not inside Leia.

Leia’s value is the knowledge it carries: the architect subagent can only emit manifests as correct as the schemas bundled with the plugin. Each release syncs that knowledge to a newer core.

The largest knowledge jump so far, covering six core releases (v0.29 → v0.36):

  • Tool-type clarity. Only builtin, agent and client load from agent YAML. The architect now offers client tools as its primary recommendation and never emits the non-loadable types (internal, webhook, rag, mcp_*) it used to suggest.
  • Model-block consumption matrix. Which keys each source actually reads — Ollama’s nested options and num_predict, timeout honoured on openai_compat, litellm ignoring endpoint. Unconsumed keys now produce a warning instead of being silently dropped.
  • Streaming contract and usage accounting. The on-wire tool_call/tool_result protocol and usage.by_model for per-model cost tracking within a run.
  • Nexus API coverage. RAG pipeline CRUD and the workflow-registration endpoint.

Taught the plugin the spec.model.default + roles shape, the litellm cloud multi-provider source, the role vocabulary each orchestration pattern requests, and orchestration.role_map. The architect can now bind a strong cloud planner to one role and a cheap local worker to another.

The Moonshot provider recipe via openai_compat, including the thinking-model behaviour and cache-aware pricing.

Leia lives inside Claude Code and drives a Nexus cluster. Its components — commands, subagents, templates, and schemas — turn your intent into manifests and API calls; Nexus fans those agents out to per-tenant astromesh-nodes, which connect to WhatsApp and to local or cloud language models.

LayerWhat it does
CommandsThe 10 /leia ... slash commands — your entry points for create, deploy, status, logs, test, and cluster lifecycle.
Agents (subagents)5 specialized subagents (interpreter, architect, operator, tester, doctor) that natural language routes to.
Templates6 complete astromesh/v1 business-vertical agent manifests you can deploy as-is or customize.
SchemasBundled references — the astromesh/v1 agent spec (including per-role models on core 0.29.0+: assign a different model/source per orchestration role such as planner/worker/synthesizer), orchestration patterns, WhatsApp config, and the Nexus API.
Nexus APIThe control plane Leia calls over REST + kubectl to provision tenants and sync agents.
astromesh-nodesThe per-tenant runtimes where agents actually execute and connect to WhatsApp, Ollama, and cloud LLMs.

Leia is the human-friendly front door to Nexus. You drive Leia; Leia drives Nexus; Nexus runs your agents on per-tenant nodes. See the ecosystem overview for how Leia, Nexus, and the nodes fit together.