flowchart TB
    subgraph USER["👤 User Layer"]
        U[User]
    end

    subgraph ENTRY["🚪 Entry Layer"]
        CLI[CLI / MCP Server]
        AID[AIDefence Security]
    end

    subgraph ROUTING["🧭 Routing Layer"]
        QL[Q-Learning Router]
        MOE[MoE - 8 Experts]
        SK[Skills - 42+]
        HK[Hooks - 17]
    end

    subgraph SWARM["🐝 Swarm Coordination"]
        TOPO[Topologies<br/>mesh/hier/ring/star]
        CONS[Consensus<br/>Raft/BFT/Gossip/CRDT]
        CLM[Claims<br/>Human-Agent Coord]
    end

    subgraph AGENTS["🤖 60+ Agents"]
        AG1[coder]
        AG2[tester]
        AG3[reviewer]
        AG4[architect]
        AG5[security]
        AG6[...]
    end

    subgraph RESOURCES["📦 Resources"]
        MEM[(Memory<br/>AgentDB)]
        PROV[Providers<br/>Claude/GPT/Gemini/Ollama]
        WORK[Workers - 12<br/>ultralearn/audit/optimize]
    end

    subgraph RUVECTOR["🧠 RuVector Intelligence Layer"]
        direction TB
        subgraph ROW1[" "]
            SONA[SONA<br/>Self-Optimize<br/>&lt;0.05ms]
            EWC[EWC++<br/>No Forgetting]
            FLASH[Flash Attention<br/>2.49-7.47x]
        end
        subgraph ROW2[" "]
            HNSW[HNSW<br/>150x-12,500x faster]
            RB[ReasoningBank<br/>Pattern Store]
            HYP[Hyperbolic<br/>Poincaré]
        end
        subgraph ROW3[" "]
            LORA[LoRA/Micro<br/>128x compress]
            QUANT[Int8 Quant<br/>3.92x memory]
            RL[9 RL Algos<br/>Q/SARSA/PPO/DQN]
        end
    end

    subgraph LEARNING["🔄 Learning Loop"]
        L1[RETRIEVE] --> L2[JUDGE] --> L3[DISTILL] --> L4[CONSOLIDATE] --> L5[ROUTE]
    end

    U --> CLI
    CLI --> AID
    AID --> QL & MOE & SK & HK
    QL & MOE & SK & HK --> TOPO & CONS & CLM
    TOPO & CONS & CLM --> AG1 & AG2 & AG3 & AG4 & AG5 & AG6
    AG1 & AG2 & AG3 & AG4 & AG5 & AG6 --> MEM & PROV & WORK
    MEM --> SONA & EWC & FLASH
    SONA & EWC & FLASH --> HNSW & RB & HYP
    HNSW & RB & HYP --> LORA & QUANT & RL
    LORA & QUANT & RL --> L1
    L5 -.->|loops back| QL

    style RUVECTOR fill:#1a1a2e,stroke:#e94560,stroke-width:2px
    style LEARNING fill:#0f3460,stroke:#e94560,stroke-width:2px
    style USER fill:#16213e,stroke:#0f3460
    style ENTRY fill:#1a1a2e,stroke:#0f3460
    style ROUTING fill:#1a1a2e,stroke:#0f3460
    style SWARM fill:#1a1a2e,stroke:#0f3460
    style AGENTS fill:#1a1a2e,stroke:#0f3460
    style RESOURCES fill:#1a1a2e,stroke:#0f3460

RuVector Components (included with OpenCode Orchestrator):

# Use RuVector via OpenCode Orchestrator
npx opencode-orchestrator@latest hooks intelligence --status

Every request flows through four layers: from your CLI or OpenCode interface, through intelligent routing, to specialized agents, and finally to LLM providers for reasoning.

Agents organize into swarms led by queens that coordinate work, prevent drift, and reach consensus on decisions—even when some agents fail.

Hive Mind Capabilities:

• 🐝 Queen Types: Strategic (planning), Tactical (execution), Adaptive (optimization)
• 👷 8 Worker Types: Researcher, Coder, Analyst, Tester, Architect, Reviewer, Optimizer, Documenter
• 🗳️ 3 Consensus Algorithms: Majority, Weighted (Queen 3x), Byzantine (f < n/3)
• 🧠 Collective Memory: Shared knowledge, LRU cache, SQLite persistence with WAL
• ⚡ Performance: Fast batch spawning with parallel agent coordination

The system stores successful patterns in vector memory, builds a knowledge graph for structural understanding, learns from outcomes via neural networks, and adapts routing based on what works best.

Skip expensive LLM calls for simple tasks using WebAssembly transforms, and compress tokens to reduce API costs by 30-50%.

Background daemons handle security audits, performance optimization, and session persistence automatically while you work.

Smart routing skips expensive LLM calls when possible. Simple edits use WASM (free), medium tasks use cheaper models. This can extend your OpenCode usage by 250% or save significantly on direct API costs.

Agent Booster uses WebAssembly to handle simple code transformations without calling the LLM at all. When the hooks system detects a simple task, it routes directly to Agent Booster for instant results.

Supported Transform Intents:

Hook Signals:

When you see these in hook output, the system is telling you how to optimize:

# Agent Booster available - skip LLM entirely
[AGENT_BOOSTER_AVAILABLE] Intent: var-to-const
→ Use Edit tool directly, 352x faster than LLM

# Model recommendation for Task tool
[TASK_MODEL_RECOMMENDATION] Use model="haiku"
→ Pass model="haiku" to Task tool for cost savings

Performance:

The Token Optimizer integrates agentic-flow optimizations to reduce API costs by compressing context and caching results.

Savings Breakdown:

Usage:

import { getTokenOptimizer } from '@claude-flow/integration';
const optimizer = await getTokenOptimizer();

// Get compact context (32% fewer tokens)
const ctx = await optimizer.getCompactContext("auth patterns");

// Optimized edit (352x faster for simple transforms)
await optimizer.optimizedEdit(file, oldStr, newStr, "typescript");

// Optimal config for swarm (100% success rate)
const config = optimizer.getOptimalConfig(agentCount);

Complex swarms can drift from their original goals. OpenCode Orchestrator V3 includes anti-drift defaults that prevent agents from going off-task.

Recommended Configuration:

// Anti-drift defaults (ALWAYS use for coding tasks)
swarm_init({
  topology: "hierarchical",  // Single coordinator enforces alignment
  maxAgents: 8,              // Smaller team = less drift surface
  strategy: "specialized"    // Clear roles reduce ambiguity
})

Why This Prevents Drift:

Additional Anti-Drift Measures:

• Frequent checkpoints via post-task hooks
• Shared memory namespace for all agents
• Short task cycles with verification gates
• Hierarchical coordinator reviews all outputs

Task → Agent Routing (Anti-Drift):

Examples:

# Minimal global install (fastest)
npm install -g opencode-orchestrator@latest --omit=optional

# With MCP auto-setup
opencode-orchestrator init --setup-mcp

# Full setup with diagnostics
curl ... | bash -s -- --full

Speed:

OpenCode Orchestrator v3 introduces self-learning neural capabilities that no other agent orchestration framework offers. While competitors require manual agent configuration and static routing, OpenCode Orchestrator learns from every task execution, prevents catastrophic forgetting of successful patterns, and intelligently routes work to specialized experts.

🧠 Neural & Learning

💾 Memory & Embeddings

🐝 Swarm & Coordination

🔧 Developer Experience

🛡️ Security & Platform

_{Comparison updated February 2026. Feature availability based on public documentation.}

What makes OpenCode Orchestrator different from other agent frameworks? These 10 capabilities work together to create a system that learns from experience, runs efficiently on any hardware, and keeps working even when things go wrong.

Not every task needs the most powerful (and expensive) model. OpenCode Orchestrator analyzes each request and automatically routes it to the cheapest handler that can do the job well. Simple code transforms skip the LLM entirely using WebAssembly. Medium tasks use faster, cheaper models. Only complex architecture decisions use Opus.

Cost & Usage Benefits:

Routing Tiers:

Benchmark Results: 100% routing accuracy, 0.57ms avg routing decision latency

Complex projects fail when implementation drifts from the original plan. OpenCode Orchestrator solves this with a spec-first approach: define your architecture through ADRs (Architecture Decision Records), organize code into DDD bounded contexts, and let the system enforce compliance as agents work. The result is implementations that match specifications — even across multi-agent swarms working in parallel.

How It Prevents Drift:

Specification Features:

DDD Bounded Contexts:

┌─────────────┐  ┌─────────────┐  ┌─────────────┐
│    Core     │  │   Memory    │  │  Security   │
│  Agents,    │  │  AgentDB,   │  │  AIDefence, │
│  Swarms,    │  │  HNSW,      │  │  Validation │
│  Tasks      │  │  Cache      │  │  CVE Fixes  │
└─────────────┘  └─────────────┘  └─────────────┘
┌─────────────┐  ┌─────────────┐
│ Integration │  │Coordination │
│ agentic-    │  │  Consensus, │
│ flow,MCP    │  │  Hive-Mind  │
└─────────────┘  └─────────────┘

Key ADRs:

• ADR-001: agentic-flow@alpha as foundation (eliminates 10,000+ duplicate lines)
• ADR-006: Unified Memory Service with AgentDB
• ADR-008: Vitest testing framework (10x faster than Jest)
• ADR-009: Hybrid Memory Backend (SQLite + HNSW)
• ADR-026: Intelligent 3-tier model routing
• ADR-048: Auto Memory Bridge (OpenCode ↔ AgentDB bidirectional sync)
• ADR-049: Self-Learning Memory with GNN (LearningBridge, MemoryGraph, AgentMemoryScope)

flowchart TB
    subgraph User["👤 User Layer"]
        CC[OpenCode]
        CLI[CLI Commands]
    end

    subgraph Orchestration["🎯 Orchestration Layer"]
        MCP[MCP Server]
        Router[Intelligent Router]
        Hooks[Self-Learning Hooks]
    end

    subgraph Agents["🤖 Agent Layer"]
        Queen[Queen Coordinator]
        Workers[60+ Specialized Agents]
        Swarm[Swarm Manager]
    end

    subgraph Intelligence["🧠 Intelligence Layer"]
        SONA[SONA Learning]
        MoE[Mixture of Experts]
        HNSW[HNSW Vector Search]
    end

    subgraph Providers["☁️ Provider Layer"]
        Anthropic[Anthropic]
        OpenAI[OpenAI]
        Google[Google]
        Ollama[Ollama]
    end

    CC --> MCP
    CLI --> MCP
    MCP --> Router
    Router --> Hooks
    Hooks --> Queen
    Queen --> Workers
    Queen --> Swarm
    Workers --> Intelligence
    Intelligence --> Providers

sequenceDiagram
    participant U as User
    participant R as Router
    participant H as Hooks
    participant A as Agent Pool
    participant M as Memory
    participant P as Provider

    U->>R: Submit Task
    R->>H: pre-task hook
    H->>H: Analyze complexity

    alt Simple Task
        H->>A: Agent Booster (WASM)
        A-->>U: Result (<1ms)
    else Medium Task
        H->>A: Spawn Haiku Agent
        A->>M: Check patterns
        M-->>A: Cached context
        A->>P: LLM Call
        P-->>A: Response
        A->>H: post-task hook
        H->>M: Store patterns
        A-->>U: Result
    else Complex Task
        H->>A: Spawn Swarm
        A->>A: Coordinate agents
        A->>P: Multiple LLM calls
        P-->>A: Responses
        A->>H: post-task hook
        A-->>U: Result
    end

flowchart LR
    subgraph Input["📥 Input"]
        Query[Query/Pattern]
        Insight[New Insight]
    end

    subgraph Processing["⚙️ Processing"]
        Embed[ONNX Embeddings]
        Normalize[Normalization]
        Learn[LearningBridge<br/>SONA + ReasoningBank]
    end

    subgraph Storage["💾 Storage"]
        HNSW[(HNSW Index<br/>150x faster)]
        SQLite[(SQLite Cache)]
        AgentDB[(AgentDB)]
        Graph[MemoryGraph<br/>PageRank + Communities]
    end

    subgraph Retrieval["🔍 Retrieval"]
        Vector[Vector Search]
        Semantic[Semantic Match]
        Rank[Graph-Aware Ranking]
        Results[Top-K Results]
    end

    Query --> Embed
    Embed --> Normalize
    Normalize --> HNSW
    Normalize --> SQLite
    Insight --> Learn
    Learn --> AgentDB
    AgentDB --> Graph
    HNSW --> Vector
    SQLite --> Vector
    AgentDB --> Semantic
    Vector --> Rank
    Semantic --> Rank
    Graph --> Rank
    Rank --> Results

Self-Learning Memory (ADR-049):

OpenCode Orchestrator V3 integrates AgentDB v3 (3.0.0-alpha.10) providing 20+ memory controllers accessible via MCP tools and the CLI.

Core Memory:

Intelligence:

Causal & Explainable:

Security & Integrity:

Optimization:

MCP Tool Examples:

# Store to hierarchical memory
agentdb_hierarchical-store --key "auth-pattern" --value "JWT refresh" --tier "semantic"

# Recall from memory tiers
agentdb_hierarchical-recall --query "authentication" --topK 5

# Run memory consolidation
agentdb_consolidate

# Batch insert
agentdb_batch --operation insert --entries '[{"key":"k1","value":"v1"}]'

# Synthesize context
agentdb_context-synthesize --query "error handling patterns"

# Semantic routing
agentdb_semantic-route --input "fix auth bug in login"

Hierarchical Memory Tiers:

┌─────────────────────────────────────────────┐
│  Working Memory                             │  ← Active context, fast access
│  Size-based eviction (1MB limit)            │
├─────────────────────────────────────────────┤
│  Episodic Memory                            │  ← Recent patterns, moderate retention
│  Importance × retention score ranking       │
├─────────────────────────────────────────────┤
│  Semantic Memory                            │  ← Consolidated knowledge, persistent
│  Promoted from episodic via consolidation   │
└─────────────────────────────────────────────┘

flowchart TB
    subgraph Hierarchical["👑 Hierarchical (Default)"]
        Q1[Queen] --> W1[Worker 1]
        Q1 --> W2[Worker 2]
        Q1 --> W3[Worker 3]
    end

    subgraph Mesh["🕸️ Mesh"]
        M1[Agent] <--> M2[Agent]
        M2 <--> M3[Agent]
        M3 <--> M1[Agent]
    end

    subgraph Ring["💍 Ring"]
        R1[Agent] --> R2[Agent]
        R2 --> R3[Agent]
        R3 --> R1
    end

    subgraph Star["⭐ Star"]
        S1[Hub] --> S2[Agent]
        S1 --> S3[Agent]
        S1 --> S4[Agent]
    end

flowchart TB
    subgraph Input["📥 Input Validation"]
        Req[Request] --> Scan[AIDefence Scan]
        Scan --> PII[PII Detection]
        Scan --> Inject[Injection Check]
        Scan --> Jailbreak[Jailbreak Detection]
    end

    subgraph Decision["⚖️ Decision"]
        PII --> Risk{Risk Level}
        Inject --> Risk
        Jailbreak --> Risk
    end

    subgraph Action["🎬 Action"]
        Risk -->|Safe| Allow[✅ Allow]
        Risk -->|Warning| Sanitize[🧹 Sanitize]
        Risk -->|Threat| Block[⛔ Block]
    end

    subgraph Learn["📚 Learning"]
        Allow --> Log[Log Pattern]
        Sanitize --> Log
        Block --> Log
        Log --> Update[Update Model]
    end

OpenCode Orchestrator runs as an MCP (Model Context Protocol) server, allowing you to connect it to any MCP-compatible AI client. This means you can use OpenCode Orchestrator's 60+ agents, swarm coordination, and self-learning capabilities from Claude Desktop, VS Code, Cursor, Windsurf, ChatGPT, and more.

Quick Add Command

# Start OpenCode Orchestrator MCP server in any environment
npx opencode-orchestrator@v3alpha mcp start

{
  "mcpServers": {
    "opencode-orchestrator": {
      "command": "npx",
      "args": ["opencode-orchestrator@v3alpha", "mcp", "start"],
      "env": {
        "ANTHROPIC_API_KEY": "sk-ant-..."
      }
    }
  }
}

# Add via CLI (recommended)
opencode mcp add opencode-orchestrator -- npx opencode-orchestrator@latest mcp start

# Or add with environment variables
opencode mcp add opencode-orchestrator \
  --env ANTHROPIC_API_KEY=sk-ant-... \
  -- npx opencode-orchestrator@latest mcp start

# Verify installation
opencode mcp list

{
  "mcpServers": {
    "opencode-orchestrator": {
      "command": "npx",
      "args": ["opencode-orchestrator@v3alpha", "mcp", "start"],
      "env": {
        "ANTHROPIC_API_KEY": "sk-ant-..."
      }
    }
  }
}

{
  "mcpServers": {
    "opencode-orchestrator": {
      "command": "npx",
      "args": ["opencode-orchestrator@v3alpha", "mcp", "start"],
      "env": {
        "ANTHROPIC_API_KEY": "sk-ant-..."
      }
    }
  }
}

{
  "mcpServers": {
    "opencode-orchestrator": {
      "command": "npx",
      "args": ["opencode-orchestrator@v3alpha", "mcp", "start"],
      "env": {
        "ANTHROPIC_API_KEY": "sk-ant-..."
      }
    }
  }
}

# Start with HTTP transport
npx opencode-orchestrator@v3alpha mcp start --transport http --port 3000

import { GoogleGenAI } from '@google/genai';

const ai = new GoogleGenAI({ apiKey: 'YOUR_API_KEY' });

// MCP definitions are natively supported in the Gen AI SDK
const mcpConfig = {
  servers: [{
    name: 'opencode-orchestrator',
    command: 'npx',
    args: ['opencode-orchestrator@v3alpha', 'mcp', 'start']
  }]
};

{
  "name": "opencode-orchestrator",
  "command": "npx",
  "args": ["opencode-orchestrator@v3alpha", "mcp", "start"]
}

Environment Variables

All configurations support these environment variables:

MCP Tool Groups

Control which MCP tools are loaded to reduce latency and token usage:

# Enable specific tool groups
export CLAUDE_FLOW_TOOL_GROUPS=implement,test,fix,memory

# Or use a preset mode
export CLAUDE_FLOW_TOOL_MODE=develop

Available Groups: create, issue, branch, implement, test, fix, optimize, monitor, security, memory, all, minimal

Preset Modes:

Precedence: CLI args (--tools=X) > Environment vars > Config file > Default (all)

Security Best Practices

⚠️ Never hardcode API keys in config files checked into version control.

# Use environment variables instead
export ANTHROPIC_API_KEY="sk-ant-..."

# Or use a .env file (add to .gitignore)
echo "ANTHROPIC_API_KEY=sk-ant-..." >> .env

Overview

@claude-flow/guidance turns CLAUDE.md into a runtime governance system with enforcement gates, cryptographic proofs, and feedback loops. Agents that normally drift after 30 minutes can now operate for days — rules are enforced mechanically at every step, not remembered by the model.

7-phase pipeline: Compile → Retrieve → Enforce → Trust → Prove → Defend → Evolve

Install

npm install @claude-flow/guidance@alpha

Quick Usage

import {
  createCompiler,
  createRetriever,
  createGates,
  createLedger,
  createProofChain,
} from '@claude-flow/guidance';

// Compile CLAUDE.md into a policy bundle
const compiler = createCompiler();
const bundle = await compiler.compile(claudeMdText);

// Retrieve task-relevant rules
const retriever = createRetriever();
await retriever.loadBundle(bundle);
const { shards, policyText } = await retriever.retrieve({
  taskDescription: 'Fix authentication bug in login flow',
});

// Enforce gates on tool calls
const gates = createGates(bundle);
const result = gates.evaluate({ tool: 'bash', args: { command: 'rm -rf /' } });
// result.blocked === true

// Audit with proof chain
const chain = createProofChain({ signingKey: process.env.PROOF_KEY! });
const envelope = chain.seal(runEvent);
chain.verify(envelope); // true — tamper-evident

Key Modules

Stats

• 1,331 tests across 26 test files
• 27 subpath exports for tree-shaking
• WASM kernel for security-critical hot paths (gates, proof, scoring)
• 25 ADRs documenting every architectural decision

Documentation

• Architecture Overview
• Getting Started
• Enforcement Gates Tutorial
• Proof Audit Trail
• Multi-Agent Security
• API Quick Reference
• Full README

Comprehensive feature set for enterprise-grade AI agent orchestration.

npx opencode-orchestrator hive-mind init                    # Initialize hive mind
npx opencode-orchestrator hive-mind spawn "Build API"       # Spawn with objective
npx opencode-orchestrator hive-mind spawn "..." --queen-type strategic --consensus byzantine
npx opencode-orchestrator hive-mind status                  # Check status
npx opencode-orchestrator hive-mind metrics                 # Performance metrics
npx opencode-orchestrator hive-mind memory                  # Collective memory stats
npx opencode-orchestrator hive-mind sessions                # List active sessions

# Automatically enabled with opencode-orchestrator init
npx opencode-orchestrator@latest init

# Or manually add to .claude/settings.json
{
  "env": {
    "CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS": "1"
  }
}

// Create a team
TeamCreate({ team_name: "feature-dev", description: "Building feature" })

// Create shared tasks
TaskCreate({ subject: "Design API", description: "..." })
TaskCreate({ subject: "Implement endpoints", description: "..." })

// Spawn teammates (parallel background work)
Task({ prompt: "Work on task #1...", subagent_type: "architect",
       team_name: "feature-dev", name: "architect", run_in_background: true })
Task({ prompt: "Work on task #2...", subagent_type: "coder",
       team_name: "feature-dev", name: "developer", run_in_background: true })

// Message teammates
SendMessage({ type: "message", recipient: "developer",
              content: "Prioritize auth", summary: "Priority update" })

// Cleanup when done
SendMessage({ type: "shutdown_request", recipient: "developer" })
TeamDelete()

# Handle idle teammate
npx opencode-orchestrator@latest hooks teammate-idle --auto-assign true

# Handle task completion
npx opencode-orchestrator@latest hooks task-completed --task-id <id> --train-patterns

# Install Quality Engineering plugin
npm install @claude-flow/plugin-agentic-qe

# Install AI Interpretability plugin
npm install @claude-flow/plugin-prime-radiant

# Install Gas Town Bridge plugin (WASM-accelerated orchestration)
npx opencode-orchestrator@latest plugins install -n @claude-flow/plugin-gastown-bridge

# Install domain-specific plugins
npm install @claude-flow/plugin-healthcare-clinical
npm install @claude-flow/plugin-financial-risk
npm install @claude-flow/plugin-legal-contracts

# Install development intelligence plugins
npm install @claude-flow/plugin-code-intelligence
npm install @claude-flow/plugin-test-intelligence
npm install @claude-flow/plugin-perf-optimizer

# Install advanced AI/reasoning plugins
npm install @claude-flow/plugin-neural-coordination
npm install @claude-flow/plugin-cognitive-kernel
npm install @claude-flow/plugin-quantum-optimizer
npm install @claude-flow/plugin-hyperbolic-reasoning

# List all installed plugins
npx opencode-orchestrator plugins list --installed

import { createRuVectorBridge } from '@claude-flow/plugins';

const bridge = createRuVectorBridge({
  host: 'localhost',
  port: 5432,
  database: 'vectors',
  user: 'postgres',
  password: 'secret',
  pool: { min: 2, max: 10 },
  ssl: true
});

// Enable the plugin
await registry.register(bridge);
await registry.loadAll();

import { createHyperbolicSpace } from '@claude-flow/plugins';

const space = createHyperbolicSpace('poincare', { curvature: -1.0 });

// Embed hierarchical data (trees, taxonomies)
const embedding = await space.embed(vector);
const distance = await space.distance(v1, v2);  // Geodesic distance
const midpoint = await space.geodesicMidpoint(v1, v2);

import { createSelfLearningSystem } from '@claude-flow/plugins';

const learning = createSelfLearningSystem(bridge);

// Automatic optimization
await learning.startLearningLoop();  // Runs in background

// Manual optimization
const suggestions = await learning.queryOptimizer.analyze(query);
await learning.indexTuner.tune('my_index');

npx opencode-orchestrator@v3alpha worker dispatch --trigger audit --context "./src"
npx opencode-orchestrator@v3alpha worker status

# Initialize ONNX embeddings with hyperbolic config
opencode-orchestrator embeddings init

# Use larger model for higher quality
opencode-orchestrator embeddings init --model all-mpnet-base-v2

# Semantic search
opencode-orchestrator embeddings search -q "authentication patterns"

# Initialize RuVector in PostgreSQL
opencode-orchestrator ruvector init --database mydb --user admin

# Check connection and schema status
opencode-orchestrator ruvector status --verbose

# Run pending migrations
opencode-orchestrator ruvector migrate --up

# Performance benchmark
opencode-orchestrator ruvector benchmark --iterations 1000

# Optimize indices and vacuum
opencode-orchestrator ruvector optimize --analyze

# Backup vector data
opencode-orchestrator ruvector backup --output ./backup.sql

npx opencode-orchestrator hive-mind init                                    # Initialize
npx opencode-orchestrator hive-mind spawn "Build API" --queen-type tactical # Spawn swarm
npx opencode-orchestrator hive-mind spawn "Research AI" --consensus byzantine --claude
npx opencode-orchestrator hive-mind status                                  # Check status

▊ OpenCode Orchestrator V3 ● ruvnet  │  ⎇ main  │  Opus 4.6  | ●42% ctx  | $0.15
🏗️ DDD [●●●●○] 4/5  ⚡ HNSW 150x  🤖 ◉ [12/8]  👥 3  🟢 CVE 3/3  💾 512MB  🧠 15%  📦 AgentDB ●1.2K vectors

{
  "statusLine": {
    "type": "command",
    "command": "node .claude/helpers/statusline.cjs"
  }
}

npx opencode-orchestrator@latest init --update --settings

{
  "model": { "display_name": "Opus 4.6" },
  "context_window": { "used_percentage": 42, "remaining_percentage": 58 },
  "cost": { "total_cost_usd": 0.15, "total_duration_ms": 45000 },
  "workspace": { "current_dir": "/path/to/project" },
  "session_id": "abc-123"
}

# Start daemon (auto-runs on SessionStart hooks)
npx opencode-orchestrator@v3alpha daemon start

# Check status with worker history
npx opencode-orchestrator@v3alpha daemon status

# Manually trigger a worker
npx opencode-orchestrator@v3alpha daemon trigger map

# Enable/disable workers
npx opencode-orchestrator@v3alpha daemon enable map audit optimize

# Stop daemon
npx opencode-orchestrator@v3alpha daemon stop

+-- Worker Daemon ---+
| Status: ● RUNNING  |
| PID: 12345         |
| Workers Enabled: 5 |
| Max Concurrent: 3  |
+--------------------+

Worker Status
+-------------+----+----------+------+---------+----------+----------+
| Worker      | On | Status   | Runs | Success | Last Run | Next Run |
+-------------+----+----------+------+---------+----------+----------+
| map         | ✓  | idle     | 12   | 100%    | 2m ago   | in 3m    |
| audit       | ✓  | idle     | 6    | 100%    | 5m ago   | in 5m    |
| optimize    | ✓  | running  | 4    | 100%    | now      | -        |
| consolidate | ✓  | idle     | 2    | 100%    | 15m ago  | in 15m   |
| testgaps    | ✓  | idle     | 3    | 100%    | 8m ago   | in 12m   |
+-------------+----+----------+------+---------+----------+----------+

# Start all daemons
.claude/helpers/daemon-manager.sh start 3 5

# Check daemon status
.claude/helpers/daemon-manager.sh status

# Stop all daemons
.claude/helpers/daemon-manager.sh stop

# Start worker manager
.claude/helpers/worker-manager.sh start 60

# Force run all workers immediately
.claude/helpers/worker-manager.sh force

# Check worker status
.claude/helpers/worker-manager.sh status

# Initialize project with wizard
npx opencode-orchestrator@v3alpha init --wizard

# Start daemon with background workers
npx opencode-orchestrator@v3alpha daemon start

# Spawn an agent with specific type
npx opencode-orchestrator@v3alpha agent spawn -t coder --name my-coder

# Initialize swarm with V3 mode
npx opencode-orchestrator@v3alpha swarm init --v3-mode

# Search memory (HNSW-indexed, 150x faster)
npx opencode-orchestrator@v3alpha memory search -q "authentication patterns"

# Run security scan
npx opencode-orchestrator@v3alpha security scan --depth full

# Performance benchmark
npx opencode-orchestrator@v3alpha performance benchmark --suite all

# Run full diagnostics
npx opencode-orchestrator@v3alpha doctor

# Run diagnostics with auto-fix
npx opencode-orchestrator@v3alpha doctor --fix

# Check specific component
npx opencode-orchestrator@v3alpha doctor --component memory

# Verbose output
npx opencode-orchestrator@v3alpha doctor --verbose

🩺 OpenCode Orchestrator Doctor v3.0.0-alpha

✅ Node.js      20.11.0 (required: 20+)
✅ npm          10.2.4 (required: 9+)
✅ Git          2.43.0
✅ Config       Valid claude-flow.config.json
✅ Daemon       Running (PID: 12345)
✅ Memory       SQLite healthy, 1.2MB
⚠️ API Keys    ANTHROPIC_API_KEY set, OPENAI_API_KEY missing
✅ MCP Server   Responsive (45ms latency)
✅ Disk Space   2.4GB available

Summary: 9/10 checks passed

# Initialize embeddings system
npx opencode-orchestrator@v3alpha embeddings init

# Generate embedding for text
npx opencode-orchestrator@v3alpha embeddings embed "authentication patterns"

# Batch embed multiple texts
npx opencode-orchestrator@v3alpha embeddings batch --file texts.txt

# Search with semantic similarity
npx opencode-orchestrator@v3alpha embeddings search "login flow" --top-k 5

import { createEmbeddingServiceAsync } from '@claude-flow/embeddings';

const service = await createEmbeddingServiceAsync({
  model: 'all-MiniLM-L6-v2',
  hyperbolic: true,  // Enable Poincaré ball embeddings
  cacheSize: 256
});

// Generate embedding
const embedding = await service.embed("authentication flow");

// Search similar patterns
const results = await service.search("login", { topK: 5 });

👨‍💻 Development & Code Quality

🔒 Security & Compliance

🐝 Multi-Agent Swarms

📊 Performance & Optimization

🔄 GitHub & DevOps

📋 Spec-Driven Development

🧠 Learning & Intelligence

The Problem

Claude Code has a finite context window (~200K tokens). When full, it compacts — summarizing the conversation and discarding details like exact file paths, tool outputs, decision reasoning, and code snippets. This creates a "context cliff" where Claude loses the ability to reference earlier work.

The Solution: Context Autopilot (ADR-051)

OpenCode Orchestrator intercepts the compaction lifecycle with three hooks that make context loss invisible:

Every Prompt                    Context Full                    After Compact
     │                              │                              │
     ▼                              ▼                              ▼
UserPromptSubmit              PreCompact                     SessionStart
     │                              │                              │
 Archive turns              Archive + BLOCK              Restore from archive
 to SQLite                  auto-compaction               via additionalContext
 (incremental)              (exit code 2)                (importance-ranked)
     │                              │                              │
     ▼                              ▼                              ▼
 Track tokens              Manual /compact               Seamless continuation
 Report % used             still allowed                 with full history

How Memory is Optimized

Optimization Thresholds

Real-Time Statusline

The statusline shows live context metrics read from autopilot-state.json:

🛡️  45% 89.2K ⊘  🧠 86%
│    │   │     │    │   │
│    │   │     │    │   └─ Intelligence score (learning.json + patterns + archive)
│    │   │     │    └──── Intelligence indicator
│    │   │     └───────── No prune cycles (⊘) or prune count (⟳N)
│    │   └─────────────── Token count (actual API usage)
│    └─────────────────── Context percentage used
└──────────────────────── Autopilot active (shield icon)

Storage Tiers

Configuration

# Context Autopilot (all have sensible defaults)
CLAUDE_FLOW_CONTEXT_AUTOPILOT=true        # Enable/disable autopilot (default: true)
CLAUDE_FLOW_CONTEXT_WINDOW=200000         # Context window size in tokens
CLAUDE_FLOW_AUTOPILOT_WARN=0.70           # Warning threshold (70%)
CLAUDE_FLOW_AUTOPILOT_PRUNE=0.85          # Optimization threshold (85%)
CLAUDE_FLOW_COMPACT_RESTORE_BUDGET=4000   # Max chars restored after compaction
CLAUDE_FLOW_RETENTION_DAYS=30             # Auto-prune never-accessed entries
CLAUDE_FLOW_AUTO_OPTIMIZE=true            # Importance ranking + pruning + sync

Commands

# Check archive status and autopilot state
node .claude/helpers/context-persistence-hook.mjs status

# Manual compact (archives first, then allows Claude Code to compress)
# Use /compact in Claude Code — autopilot allows manual, blocks auto

# Query archive directly
sqlite3 .claude-flow/data/transcript-archive.db \
  "SELECT COUNT(*), SUM(LENGTH(content)) FROM transcript_entries;"

Architecture Reference

• ADR-051: Infinite Context via Compaction-to-Memory Bridge
• ADR-052: Statusline Observability System
• Implementation: .claude/helpers/context-persistence-hook.mjs (~1560 lines)
• Settings: .claude/settings.json (PreCompact, SessionStart, UserPromptSubmit hooks)

Why RVF?

Previous versions shipped sql.js (18MB WASM blob) for persistent storage. This caused slow cold starts, large installs, and compatibility issues on ARM/Alpine. RVF eliminates all of that:

How it works

RVF files use a simple binary layout: a 4-byte magic header (RVF\0), a JSON metadata section, then packed entries. Each module has its own format variant:

Storage auto-selection

You don't need to pick a backend. The DatabaseProvider tries each option in order and uses the first one available:

RVF (pure TypeScript) → better-sqlite3 (native) → sql.js (WASM) → JSON (fallback)

RVF is always available since it has zero dependencies, so it wins by default. If you have better-sqlite3 installed (e.g., for advanced queries), it gets priority.

Vector search with HnswLite

RVF includes HnswLite — a pure TypeScript implementation of the HNSW (Hierarchical Navigable Small World) algorithm for fast nearest-neighbor search. It's used automatically when storing entries with embeddings.

import { RvfBackend } from '@claude-flow/memory';

const backend = new RvfBackend({ databasePath: './memory.rvf' });
await backend.initialize();

// Store entries — embeddings are indexed automatically
await backend.store({ id: '1', key: 'auth-pattern', content: '...', embedding: vector });

// Search by similarity
const results = await backend.search({ embedding: queryVector, limit: 10 });

Supports cosine, dot product, and Euclidean distance metrics. For large datasets (100K+ entries), install hnswlib-node for the native implementation — the backend switches automatically.

Migrating from older formats

The RvfMigrator converts between JSON files, SQLite databases, and RVF:

import { RvfMigrator } from '@claude-flow/memory';

// Auto-detect format and migrate
await RvfMigrator.autoMigrate('./old-memory.db', './memory.rvf');

// Or be explicit
await RvfMigrator.fromJsonFile('./backup.json', './memory.rvf');
await RvfMigrator.fromSqlite('./legacy.db', './memory.rvf');

// Export back to JSON for inspection
await RvfMigrator.toJsonFile('./memory.rvf', './export.json');

Format detection works by reading the first few bytes of the file — no file extension guessing.

Crash safety

All write operations use atomic writes: data goes to a temporary file first, then a single rename() call swaps it into place. If the process crashes mid-write, the old file stays intact.

• Memory backend: file.rvf.tmp → file.rvf
• Embedding cache: file.rvec.tmp.{random} → file.rvec
• Event log: Append-only (no overwrite needed)

SONA learning persistence

The PersistentSonaCoordinator stores learning patterns and trajectories in RVF format, so agents retain knowledge across sessions:

import { PersistentSonaCoordinator } from '@claude-flow/memory';

const sona = new PersistentSonaCoordinator({
  storePath: './data/sona-learning.rvls',
});
await sona.initialize();

// Patterns survive restarts
const similar = sona.findSimilarPatterns(embedding, 5);
sona.storePattern('routing', embedding);
await sona.shutdown(); // persists to disk

Security

RVF validates inputs at every boundary:

• Path validation — null bytes and traversal attempts are rejected
• Header validation — corrupted files are detected before parsing
• Payload limits — event log entries cap at 100MB to prevent memory exhaustion
• Dimension validation — embedding dimensions must be between 1 and 10,000
• Concurrent write protection — a lock flag prevents overlapping disk flushes

Configuration

# Environment variables
CLAUDE_FLOW_MEMORY_BACKEND=hybrid   # auto-selects RVF
CLAUDE_FLOW_MEMORY_PATH=./data/memory

# Or via CLI
opencode-orchestrator memory init --force
opencode-orchestrator config set memory.backend hybrid

What Are Hooks?

Hooks intercept operations (file edits, commands, tasks) and learn from outcomes. Unlike static automation, hooks improve over time by tracking what works and applying those patterns to future tasks.

How Hooks Learn (4-Step Pipeline)

┌─────────────┐    ┌─────────────┐    ┌─────────────┐    ┌─────────────┐
│  RETRIEVE   │───▶│    JUDGE    │───▶│   DISTILL   │───▶│ CONSOLIDATE │
│             │    │             │    │             │    │             │
│ Find similar│    │ Was it      │    │ Extract key │    │ Prevent     │
│ past patterns│   │ successful? │    │ learnings   │    │ forgetting  │
└─────────────┘    └─────────────┘    └─────────────┘    └─────────────┘
     HNSW              Verdict            LoRA              EWC++
   150x faster        success/fail      compression       memory lock

Hook Signals (ADR-026 Model Routing)

When hooks run, they emit signals that guide routing decisions. Watch for these in hook output:

Agent Booster Intents (handled without LLM):

• var-to-const - Convert var/let to const
• add-types - Add TypeScript type annotations
• add-error-handling - Wrap in try/catch
• async-await - Convert promises to async/await
• add-logging - Add console.log statements
• remove-console - Strip console.* calls

Example Hook Output:

$ npx opencode-orchestrator@v3alpha hooks pre-task --description "convert var to const in utils.ts"

[AGENT_BOOSTER_AVAILABLE] Intent: var-to-const
Recommendation: Use Edit tool directly
Performance: <1ms (352x faster than LLM)
Cost: $0

Intelligence Loop (ADR-050)

The intelligence loop wires PageRank-ranked memory into the hook system. Every session builds a knowledge graph that improves over time:

SessionStart:
  session-restore  → intelligence.init()
    → Read MEMORY.md / auto-memory-store.json
    → Build graph (nodes + similarity/temporal edges)
    → Compute PageRank
    → "[INTELLIGENCE] Loaded 13 patterns, 12 edges"

UserPrompt:
  route            → intelligence.getContext(prompt)
    → Jaccard-match prompt against pre-ranked entries
    → Inject top-5 patterns into Claude's context:

    [INTELLIGENCE] Relevant patterns for this task:
      * (0.95) HNSW gives 150x-12,500x speedup [rank #1, 12x accessed]
      * (0.88) London School TDD preferred [rank #3, 8x accessed]

PostEdit:
  post-edit        → intelligence.recordEdit(file)
    → Append to pending-insights.jsonl (<2ms)

SessionEnd:
  session-end      → intelligence.consolidate()
    → Process pending insights (3+ edits → new entry)
    → Confidence boost for accessed patterns (+0.03)
    → Confidence decay for unused patterns (-0.005/day)
    → Recompute PageRank, rebuild edges
    → Save snapshot for trend tracking

Measuring improvement:

# Human-readable diagnostics
node .claude/helpers/hook-handler.cjs stats

# JSON output for scripting
node .claude/helpers/hook-handler.cjs stats --json

# Or via intelligence.cjs directly
node .claude/helpers/intelligence.cjs stats

The stats command shows:

Example output:

+--------------------------------------------------------------+
|  Intelligence Diagnostics (ADR-050)                          |
+--------------------------------------------------------------+

  Graph
    Nodes:    9
    Edges:    8 (7 temporal, 1 similar)
    Density:  22.2%

  Confidence
    Min:      0.490    Max:  0.600
    Mean:     0.556    Median: 0.580

  Access
    Total accesses:     11
    Patterns used:      6/9
    Never accessed:     3

  Top Patterns (by composite score)
    #1  HNSW gives 150x-12,500x speedup
         conf=0.600  pr=0.2099  score=0.3659  accessed=2x
    #2  London School TDD preferred
         conf=0.600  pr=0.1995  score=0.3597  accessed=2x

  Last Delta (5m ago)
    Confidence: +0.0300
    Accesses:   +6

  Trend (3 snapshots)
    Confidence drift:  +0.0422
    Direction:         IMPROVING
+--------------------------------------------------------------+

All 27 Hooks by Category

🔧 Tool Lifecycle Hooks (6 hooks)

# Example: Edit with pattern learning
npx opencode-orchestrator@v3alpha hooks pre-edit ./src/auth.ts
npx opencode-orchestrator@v3alpha hooks post-edit ./src/auth.ts --success true --train-patterns

🧠 Intelligence & Routing Hooks (8 hooks)

# Route a task with explanation
npx opencode-orchestrator@v3alpha hooks route "refactor authentication to use JWT" --include-explanation

# Bootstrap intelligence from your codebase
npx opencode-orchestrator@v3alpha hooks pretrain --depth deep --model-type moe

📅 Session Management Hooks (4 hooks)

# Start session with auto-daemon
npx opencode-orchestrator@v3alpha hooks session-start --session-id "feature-auth" --start-daemon

# End session and export learnings
npx opencode-orchestrator@v3alpha hooks session-end --export-metrics --persist-patterns

🤖 Intelligence System Hooks (9 hooks)

# Start trajectory for complex task
npx opencode-orchestrator@v3alpha hooks intelligence trajectory-start --task "implement OAuth2"

# Record successful action
npx opencode-orchestrator@v3alpha hooks intelligence trajectory-step --action "created token service" --quality 0.9

# End trajectory and trigger learning
npx opencode-orchestrator@v3alpha hooks intelligence trajectory-end --success true

# View intelligence diagnostics and improvement trends (ADR-050)
node .claude/helpers/hook-handler.cjs stats
node .claude/helpers/intelligence.cjs stats --json

12 Background Workers (Auto-Triggered)

Workers run automatically based on context, or dispatch manually.

# List all workers
npx opencode-orchestrator@v3alpha hooks worker list

# Manually dispatch security audit
npx opencode-orchestrator@v3alpha hooks worker dispatch --trigger audit --context "./src/auth"

# Check worker status
npx opencode-orchestrator@v3alpha hooks worker status

Model Routing Hooks (3 hooks)

Automatically selects haiku/sonnet/opus based on task complexity.

# Get model recommendation
npx opencode-orchestrator@v3alpha hooks model-route --task "fix typo in README"
# → Recommends: haiku (simple task, low complexity)

npx opencode-orchestrator@v3alpha hooks model-route --task "design distributed consensus system"
# → Recommends: opus (complex architecture, high reasoning)

Progress Tracking

Quick Reference

# ══════════════════════════════════════════════════════════════════
# MOST COMMON HOOKS
# ══════════════════════════════════════════════════════════════════

# Route task to best agent (with intelligence context injection)
npx opencode-orchestrator@v3alpha hooks route "<task>" --include-explanation

# Start/end session with learning
npx opencode-orchestrator@v3alpha hooks session-start --start-daemon
npx opencode-orchestrator@v3alpha hooks session-end --persist-patterns

# View what the system has learned
npx opencode-orchestrator@v3alpha hooks metrics
npx opencode-orchestrator@v3alpha hooks intelligence stats

# Intelligence diagnostics — see if intelligence is improving
node .claude/helpers/hook-handler.cjs stats          # Human-readable
node .claude/helpers/hook-handler.cjs stats --json   # JSON for scripting
node .claude/helpers/intelligence.cjs stats           # Direct access

# Bootstrap on new project
npx opencode-orchestrator@v3alpha hooks pretrain --depth deep

# Dispatch background worker
npx opencode-orchestrator@v3alpha hooks worker dispatch --trigger audit

Share learned patterns across projects, teams, and the community via the decentralized pattern marketplace.

What You Can Share

Export Commands

# Export learned patterns to file
npx opencode-orchestrator@v3alpha memory export --format json --output ./patterns.json

# Export specific namespace
npx opencode-orchestrator@v3alpha memory export --namespace "security" --output ./security-patterns.json

# Export with embeddings (larger file, faster import)
npx opencode-orchestrator@v3alpha memory export --include-embeddings --output ./full-export.json

# Export agent configurations
npx opencode-orchestrator@v3alpha config export --scope project --output ./agent-configs.json

# Export session state
npx opencode-orchestrator@v3alpha session export --session-id "my-session" --output ./session.json

Import Commands

# Import patterns from file
npx opencode-orchestrator@v3alpha memory import --input ./patterns.json

# Import and merge with existing (don't overwrite)
npx opencode-orchestrator@v3alpha memory import --input ./patterns.json --merge

# Import from another project
npx opencode-orchestrator@v3alpha hooks transfer --source-path ../other-project

# Import agent configurations
npx opencode-orchestrator@v3alpha config import --input ./agent-configs.json --scope project

# Restore session
npx opencode-orchestrator@v3alpha session restore --session-id "my-session"

Pattern Store (IPFS Marketplace)

Decentralized pattern marketplace for sharing and discovering community patterns.

# Search for authentication patterns
npx opencode-orchestrator@v3alpha transfer-store search --query "authentication" --min-rating 4.0

# Download a pattern
npx opencode-orchestrator@v3alpha transfer-store download --id "auth-jwt-patterns-v2" --verify

# Publish your patterns
npx opencode-orchestrator@v3alpha transfer-store publish --input ./my-patterns.json --category "security"

Plugin Store

Discover and install community plugins from the live IPFS registry with 19 official plugins and live ratings via Cloud Function.

# List plugins with live ratings from Cloud Function
npx opencode-orchestrator@v3alpha plugins list

# Filter by type
npx opencode-orchestrator@v3alpha plugins list --type integration

# Rate a plugin
npx opencode-orchestrator@v3alpha plugins rate --name @claude-flow/embeddings --rating 5

# Search for MCP tool plugins
npx opencode-orchestrator@v3alpha transfer plugin-search --type "mcp-tool" --verified

# Get plugin info
npx opencode-orchestrator@v3alpha transfer plugin-info --name "semantic-code-search"

# List official plugins
npx opencode-orchestrator@v3alpha transfer plugin-official

Live IPFS Plugin Registry

The official plugin registry is hosted on IPFS with Ed25519 signature verification:

# Fetch live registry directly
curl -s "https://gateway.pinata.cloud/ipfs/bafkreiahw4ufxwycbwwswt7rgbx6hkgnvg3rophhocatgec4bu5e7tzk2a"

IPFS Integration

Patterns and models are distributed via IPFS for decentralization and integrity.

# Resolve IPNS name to CID
npx opencode-orchestrator@v3alpha transfer ipfs-resolve --name "/ipns/patterns.opencode-orchestrator.io"

# Detect PII before publishing
npx opencode-orchestrator@v3alpha transfer detect-pii --content "$(cat ./patterns.json)"

Model & Learning Pattern Import/Export

Share trained neural patterns and learning models via IPFS.

# Export a learning pattern to IPFS
curl -X POST "https://api.pinata.cloud/pinning/pinJSONToIPFS" \
  -H "Authorization: Bearer $PINATA_JWT" \
  -d '{
    "pinataContent": {
      "type": "learning-pattern",
      "name": "my-patterns",
      "patterns": [...]
    },
    "pinataMetadata": {"name": "opencode-orchestrator-learning-pattern"}
  }'

# Import a pattern from IPFS CID
curl -s "https://gateway.pinata.cloud/ipfs/QmYourCIDHere"

# Via Cloud Function (when deployed)
curl "https://publish-registry-xxx.cloudfunctions.net?action=export-model" -d @model.json
curl "https://publish-registry-xxx.cloudfunctions.net?action=import-model&cid=QmXxx"

Supported Model Types

Pre-trained Model Registry

Import pre-trained learning patterns for common tasks. 90.5% average accuracy across 40 patterns trained on 110,600+ examples.

Registry CID: QmNr1yYMKi7YBaL8JSztQyuB5ZUaTdRMLxJC1pBpGbjsTc

# Browse available models
curl -s "https://gateway.pinata.cloud/ipfs/QmNr1yYMKi7YBaL8JSztQyuB5ZUaTdRMLxJC1pBpGbjsTc" | jq '.models[].name'

# Import all models
npx opencode-orchestrator@v3alpha transfer import --cid QmNr1yYMKi7YBaL8JSztQyuB5ZUaTdRMLxJC1pBpGbjsTc

# Import specific category
npx opencode-orchestrator@v3alpha neural import --model security-review-patterns --source ipfs

# Use patterns in routing
npx opencode-orchestrator@v3alpha hooks route --task "review authentication code" --use-patterns

Benefits vs Fresh Install

Pre-Built Pattern Packs

# Install a pattern pack
npx opencode-orchestrator@v3alpha transfer-store download --id "security-essentials" --apply

RuVector WASM Neural Training

Real WASM-accelerated neural training using @ruvector/learning-wasm and @ruvector/attention packages for state-of-the-art performance.

# List available pre-trained models from IPFS registry
npx opencode-orchestrator@v3alpha neural list

# List models by category
npx opencode-orchestrator@v3alpha neural list --category security

# Train with WASM acceleration
npx opencode-orchestrator@v3alpha neural train -p coordination -e 100 --wasm --flash --contrastive

# Train security patterns
npx opencode-orchestrator@v3alpha neural train -p security --wasm --contrastive

# Benchmark WASM performance
npx opencode-orchestrator@v3alpha neural benchmark -d 256 -i 1000

# Import pre-trained models
npx opencode-orchestrator@v3alpha neural import --cid QmNr1yYMKi7YBaL8JSztQyuB5ZUaTdRMLxJC1pBpGbjsTc

# Export trained patterns to IPFS
npx opencode-orchestrator@v3alpha neural export --ipfs --sign

Benchmark Results

+---------------------+---------------+-------------+
| Mechanism           | Avg Time (ms) | Ops/sec     |
+---------------------+---------------+-------------+
| DotProduct          | 0.1063        | 9,410       |
| FlashAttention      | 0.1096        | 9,127       |
| MultiHead (4 heads) | 0.1661        | 6,020       |
| MicroLoRA           | 0.0026        | 383,901     |
+---------------------+---------------+-------------+
MicroLoRA Target (<100μs): ✓ PASS (2.60μs actual)

Training Options