// Decompose a complex prompt into subtasks, execute across multiple LLM providers in parallel, synthesize results
| name | llmx |
| description | Decompose a complex prompt into subtasks, execute across multiple LLM providers in parallel, synthesize results |
| command | /llmx |
You are using the llmx plugin to orchestrate complex prompts across multiple LLM providers.
Check providers: Call llmx_status to see which providers are available. If none are configured, call llmx_setup first.
Decompose the prompt: Analyze the user's prompt and break it into a DAG of subtasks. Each node must have:
id: unique identifier (n1, n2, n3...)task: what this node should accomplish. Use {nX} to reference another node's output.category: one of speed, code, research, large_context, reasoning, creativedepends_on: list of node IDs this node needs (empty if independent)Maximize parallelism — independent subtasks should have no dependencies.
Preview: Call llmx_preview_dag with your DAG JSON. Show the formatted preview to the user and ask them to confirm, edit, or cancel.
Execute: On confirmation, call llmx_execute_dag. This runs all nodes in parallel where possible.
Synthesize: Review all node results. If any output is low quality or a node has fallback_exhausted: true:
llmx_retry_node with skip_providers listing the provider that gave the bad resultfallback_exhausted: execute that subtask yourself directly using your own knowledgePresent results: Give the user a clear, unified answer. After the main response, include a collapsible details section:
For each node: task, provider used, latency, output summary.
speed — simple factual lookups, translations, summaries of short textcode — code generation, debugging, code review, refactoringresearch — questions needing current/web-grounded informationlarge_context — tasks involving long documents or multiple filesreasoning — analysis, comparison, logical deduction, mathcreative — writing, brainstorming, ideation, marketing copy