opencode-supermemory-max

🧠 Enhanced fork of opencode-supermemory — maximizing memory quality and coverage for OpenCode coding agents.

Why "max"?

The official opencode-supermemory plugin is maintained by the Supermemory team, but tends to be conservative with API calls and slow to merge fixes. opencode-supermemory-max takes the opposite approach:

• Aggressive memory utilization — save more, recall more, lose nothing
• Bug fixes shipped fast — no 2-week PR review waits
• Feature parity with claude-supermemory — all the good stuff from the Claude Code plugin, ported to OpenCode
• Japanese language support — keyword detection and signal extraction work in both English and Japanese
• Upstream tracking — good changes from upstream get merged in

What's different from upstream?

This plugin combines the best features from all three official Supermemory plugins:

• opencode-supermemory (OpenCode plugin — base)
• claude-supermemory (Claude Code plugin — entity context, dedup, signal extraction, repo tags)
• openclaw-supermemory (OpenClaw plugin — incremental capture, metadata stripping, compaction hooks)

| Feature | upstream | claude | openclaw | max | |---------|----------|--------|----------|---------| | Part ID fix (OpenCode v1.2.25+) | ❌ | — | — | ✅ | | Japanese keywords (覚えて, メモして) | ❌ | ❌ | ❌ | ✅ | | Incremental capture (crash-safe) | ❌ | ❌ | ✅ | ✅ | | Every-message recall | ❌ | ❌ | ✅ | ✅ | | Metadata stripping for queries | ❌ | ❌ | ✅ | ✅ | | Pre-compaction full save | ❌ | ❌ | ✅ | ✅ | | Post-compaction memory re-injection | ❌ | ❌ | ❌ | ✅ | | Entity Context (extraction guidance) | ❌ | ✅ | ❌ | ✅ | | Dedup (no duplicate memories) | ❌ | ✅ | ❌ | ✅ | | Relative time display | ❌ | ✅ | ❌ | ✅ | | Repo container tag (team sharing) | ❌ | ✅ | ❌ | ✅ | | Signal extraction (smart filtering) | ❌ | ✅ | ❌ | ✅ | | Context re-injection (periodic) | ❌ | ❌ | ❌ | ✅ | | Recall keywords (思い出して, recall) | ❌ | ❌ | ❌ | ✅ | | Session-end auto-save | ❌ | ✅ | ✅ | ✅ |

Installation

Quick Start

# Clone and build
git clone https://github.com/kandotrun/opencode-supermemory-max.git
cd opencode-supermemory-max
bun install && bun run build

Add to your OpenCode config (~/.config/opencode/opencode.json):

{
  "plugin": ["~/opencode-supermemory-max"]
}

Set your API key:

export SUPERMEMORY_API_KEY="sm_..."

Or create ~/.config/opencode/supermemory.jsonc:

{
  "apiKey": "sm_..."
}

Get your API key from app.supermemory.ai.

Features

🔄 Context Injection

On first message (and periodically), the agent receives memory context:

[SUPERMEMORY]

User Profile:
- Prefers concise responses
- Expert in TypeScript

Repo Knowledge (Shared):
- [2hrs ago] Uses monorepo with turborepo [85%]

Project Knowledge:
- [just now] Build: bun run build [100%]

Relevant Memories:
- [3d ago] Build fails if .env.local missing [82%]

🇯🇵 Japanese + English Keywords

Memory triggers work in both languages:

| English | Japanese | |---------|----------| | remember, save this | 覚えて, メモして, 保存して | | recall, check memory | 思い出して, メモリ確認 |

📦 3-Tier Memory Scopes

| Scope | Purpose | Tag Format | |-------|---------|------------| | User | Cross-project preferences | opencode_user_{hash} | | Project | Directory-specific knowledge | opencode_project_{hash} | | Repo | Team-shared via git remote | repo_{reponame} |

🎯 Signal Extraction

When enabled, only saves session turns containing important keywords — no noise:

{
  "signalExtraction": true,
  "signalTurnsBefore": 3  // context turns before each signal
}

Built-in signal keywords: implementation, architecture, decision, bug, fix, 実装, 設計, 重要, バグ, 修正 ...

🧹 Dedup & Entity Context

• Dedup: Duplicate memories are automatically filtered out from search results and profile
• Entity Context: Guides Supermemory on what to extract — user actions & decisions for personal scope, architecture & patterns for repo scope

⏰ Context Re-injection

{
  "reinjectEveryN": 10  // re-inject context every 10 messages
}

Or say "recall" / "思い出して" to trigger immediate context refresh.

Configuration

~/.config/opencode/supermemory.jsonc:

{
  // Memory retrieval
  "similarityThreshold": 0.6,
  "maxMemories": 5,
  "maxProjectMemories": 10,
  "maxRepoMemories": 5,
  "maxProfileItems": 5,

  // Container tags (auto-generated if not set)
  "containerTagPrefix": "opencode",
  "userContainerTag": "my-custom-user-tag",
  "projectContainerTag": "my-project-tag",
  "repoContainerTag": "my-repo-tag",

  // Context re-injection
  "reinjectEveryN": 10,

  // Signal extraction (disabled by default)
  "signalExtraction": false,
  "signalTurnsBefore": 3,

  // Compaction
  "compactionThreshold": 0.8,

  // Extra keyword patterns (regex)
  "keywordPatterns": ["log\\s+this"],
  "recallKeywordPatterns": ["my custom recall"],
  "signalKeywords": ["custom signal"]
}

Tool Usage

The supermemory tool is available to the agent:

| Mode | Args | Description | |------|------|-------------| | add | content, type?, scope? | Store memory | | search | query, scope? | Search memories | | profile | query? | View user profile | | list | scope?, limit? | List memories | | forget | memoryId, scope? | Delete memory |

Scopes: user, project (default), repo

Privacy

Content in <private> tags is never stored:

API key is <private>sk-abc123</private>

Development

bun install
bun run build
bun run typecheck

Upstream Sync

This fork tracks supermemoryai/opencode-supermemory:

git fetch upstream
git merge upstream/main

License

MIT