Skip to content

API.md

Latest commit

 

History

History
200 lines (153 loc) · 6.6 KB

API.md

File metadata and controls

200 lines (153 loc) · 6.6 KB
Raw

REPLOID API Documentation

This document is a current module map for the live Reploid tree. It is intentionally less exhaustive than older versions: source files and self/config/module-registry.json are the source of truth for exact wiring.

Source of Truth

Use these files first:

  • self/config/module-registry.json for runtime module identity and dependencies
  • self/config/genesis-levels.json for level composition
  • self/core/, self/infrastructure/, self/capabilities/, self/testing/arena/, and self/ui/ for implementation
  • self/blueprints/ for architectural intent

Module Conventions

Most runtime modules follow the standard metadata/factory pattern:

const ExampleModule = {
  metadata: {
    id: 'ExampleModule',
    version: '1.0.0',
    genesis: { introduced: 'spark' },
    dependencies: ['Utils'],
    async: true,
    type: 'core'
  },

  factory: (deps) => {
    return {
      init() {
        return true;
      }
    };
  }
};

export default ExampleModule;

Tools are file-based modules under self/tools/ and typically export:

export default async function(args, deps) {
  return result;
}

export const schema = {
  name: 'ToolName',
  description: 'What the tool does',
  parameters: {}
};

Core Runtime Modules

Module Path Purpose
Utils self/core/utils.js Error classes, logging, helpers
StateManager self/core/state-manager.js Session and artifact state
PersonaManager self/core/persona-manager.js System prompt composition, persona overrides
AgentLoop self/core/agent-loop.js Main think -> act -> observe loop
VFS self/core/vfs.js IndexedDB-backed virtual filesystem
ToolRunner self/core/tool-runner.js Tool discovery, registration, execution
WorkerManager self/core/worker-manager.js Worker orchestration and tool filtering
LLMClient self/core/llm-client.js Model-provider abstraction
ResponseParser self/core/response-parser.js Tool-call and JSON response parsing
SchemaRegistry self/core/schema-registry.js Runtime schema validation
VerificationManager self/core/verification-manager.js Pre-write validation and policy checks
MultiModelEvaluator self/core/multi-model-evaluator.js Task-suite evaluation across models
FunctionGemmaOrchestrator self/core/functiongemma-orchestrator.js Multi-expert orchestration surface

Infrastructure Modules

Module Path Purpose
EventBus self/infrastructure/event-bus.js Pub/sub for runtime coordination
DIContainer self/infrastructure/di-container.js Dependency resolution
HITLController self/infrastructure/hitl-controller.js Approval-mode control
AuditLogger self/infrastructure/audit-logger.js Audit log persistence
CircuitBreaker self/infrastructure/circuit-breaker.js Failure throttling and isolation
RateLimiter self/infrastructure/rate-limiter.js Request pacing
TraceStore self/infrastructure/trace-store.js Structured execution traces
BrowserAPIs self/infrastructure/browser-apis.js File System Access, notifications, storage, clipboard, wake lock

Capability Modules

Module Path Purpose
GEPAOptimizer self/capabilities/cognition/gepa-optimizer.js Prompt evolution and candidate scoring
SemanticMemory self/capabilities/cognition/semantic/semantic-memory-llm.js Semantic retrieval
KnowledgeGraph self/capabilities/cognition/symbolic/knowledge-graph.js Symbolic entity and relationship storage
RuleEngine self/capabilities/cognition/symbolic/rule-engine.js Deterministic symbolic reasoning
ReflectionStore self/capabilities/reflection/reflection-store.js Persistent learning and reflection storage
PerformanceMonitor self/capabilities/performance/performance-monitor.js Metrics collection and reporting
IntentBundleLoRA self/capabilities/intelligence/intent-bundle-lora.js Intent-bundle gated LoRA activation

Experimental Modules

These modules are real code but remain outside the main stable capability tree:

Module Path Purpose
NeuralCompiler self/experimental/intelligence/neural-compiler.js LoRA routing and adapter scheduling

Treat experimental paths as implementation details until they move into the main capability tree.

UI Modules

Module Path Purpose
Proto UI self/ui/proto/index.js Main operator UI
UIManager self/ui/dashboard/ui-manager.js Dashboard orchestration
VFSExplorer self/ui/dashboard/vfs-explorer.js File tree UI

Tool Surface

Current built-in tool files under self/tools/:

  • AwaitWorkers.js
  • CreateTool.js
  • DeleteFile.js
  • EditFile.js
  • FileOutline.js
  • Find.js
  • Grep.js
  • Head.js
  • ListFiles.js
  • ListKnowledge.js
  • ListMemories.js
  • ListTools.js
  • ListWorkers.js
  • LoadModule.js
  • ReadFile.js
  • RunGEPA.js
  • SpawnWorker.js
  • SwarmGetStatus.js
  • SwarmListPeers.js
  • SwarmRequestFile.js
  • SwarmShareFile.js
  • Tail.js
  • WriteFile.js
  • git.js

At runtime, tool visibility is filtered by genesis level and worker type.

Persistence Paths

Important runtime storage paths:

Path Purpose
/.memory/persona-overrides.json Persona overrides
/.memory/multi-model-eval/ Multi-model evaluation runs
/.memory/neural-compiler/adapters.json Adapter registry
/.memory/traces/ Structured trace sessions
/.logs/audit/ Audit logs
/.system/intent-bundle.json Default intent bundle path

Status Notes

  • FunctionGemmaOrchestrator is a live runtime module at self/core/functiongemma-orchestrator.js.
  • NeuralCompiler currently lives under self/experimental/intelligence/, not self/capabilities/intelligence/.
  • There is no standalone runtime module at self/infrastructure/introspector.js.
  • There is no standalone runtime module at self/core/sentinel-fsm.js; the current reference is the blueprint self/blueprints/0x000050-sentinel-fsm.md.
  • Older docs may refer to ui/diff-generator.js; the current UI surface is centered on self/ui/proto/ and self/ui/dashboard/.

Related Docs

Last updated: March 2026