Shared from "Study" on Inkdown
The orchestration layer is the brain of the Arcane system. It manages multi-agent workflows, handles iterative tool calling, maintains conversation state, and coordinates all AI-driven operations.
File: src/server/endpoints/unified/orchestrator/toolOrchestrator.ts:69
Purpose:
File: src/server/endpoints/unified/orchestrator/toolOrchestrator.ts:237
This is where the magic happens - iterative reasoning with tools:
Why This Loop Matters:
File: src/server/endpoints/unified/orchestrator/toolOrchestrator.ts:129
Critical Field: currentContentIndex
This prevents streaming conflicts in multi-agent scenarios. Each chunk sent to client has an index:
Location: src/server/endpoints/unified/orchestrator/configs/
Three Built-in Agents:
MainThreadAgent (mainThread.config.ts)
DeepResearchSupervisor (deepResearch.config.ts)
ResearcherAgent (researcher.config.ts)
File: src/server/endpoints/unified/orchestrator/toolOrchestrator.ts:929
Tools run in parallel, results stream as they arrive. No waiting for slow tools.
File: src/server/endpoints/unified/orchestrator/toolOrchestrator.ts:962
Tool Result Types:
tool:start - Execution beginstool:stream - Streaming partial results (sub-agents)tool:done - Success with final resulttool:error - Failed (caught gracefully)tool:progress - Overall batch trackingSub-Agent Spawning Example:
Engine (src/server/repositories/engine/engine.ts)
Streamer (src/server/repositories/streamer/streamer.ts)
Provider (src/server/repositories/provider/provider.ts)
ToolRegistry (src/server/endpoints/unified/tools/toolRegistry.ts)
The orchestration system:
Key insight: This isn't single-pass Q&A. It's an iterative reasoning engine where the LLM can take multiple actions, see results, and continue thinking - just like a human problem-solving approach.