Architecture Overview: Arachne

April 26, 2026 ยท View on GitHub

Arachne is a production-grade, self-healing AI agent runtime that executes autonomous tasks via dynamically generated DSPy-native graph topologies.

Core Philosophy

  • Thin Orchestration, Thick Intelligence: Minimal structural scaffolding; all "intelligence" lives in DSPy-native modules (Weaver, Evolver, Evaluator).
  • Silent Execution with Active Enrichment: Agents run autonomously, discovering missing data via tool-use and context injection.
  • Self-Healing First, HITL Last: Automatic detection of failures and re-weaving/re-routing before human intervention.
  • Stateful Persistence: Every wave and node result is checkpointed to disk for reliability and resume-ability.

System Components

Arachne is composed of several key modules working together:

  1. Graph Weaver: Generates the Directed Acyclic Graph (DAG) for a given goal.
  2. Graph Runner: Orchestrates the execution of nodes in waves.
  3. Triangulated Evaluator: Verifies the quality and correctness of results.
  4. AutoHealer: Diagnoses failures and applies recovery strategies.
  5. MCP Manager: Integrates with Model Context Protocol servers for dynamic tool discovery.

Execution Flow

graph TD
    A[Goal Submission] --> B[Graph Weaver]
    B --> C[Provisioning & Tool Injection]
    C --> D[Graph Runner]
    D --> E{Waves Success?}
    E -- No --> F[AutoHealer]
    F -- Fix Proposed --> B
    F -- Human Required --> G[HITL Pause]
    E -- Yes --> H[Triangulated Evaluator]
    H -- Pass --> I[Success]
    H -- Fail --> F