![\"SynaBun\"](\"public/synabun.png\")
\n\n \n
\n Persistent vector memory for AI assistants — powered by SQLite and local embeddings.\n
\n\n\n ![\"npm](\"https://img.shields.io/npm/v/synabun?color=cb3837&logo=npm&logoColor=white\")
![\"npm](\"https://img.shields.io/npm/dm/synabun?color=cb3837&logo=npm&logoColor=white\")
![\"license\"](\"https://img.shields.io/npm/l/synabun?color=blue\")
![\"CI\"](\"https://github.com/danilokhury/Synabun/actions/workflows/ci.yml/badge.svg\"){kind=icon}
\n ![\"Discord\"](\"https://img.shields.io/badge/Discord-Join-5865F2?logo=discord&logoColor=white\"){kind=icon}
![\"X](\"https://img.shields.io/badge/Follow-%40SynabunAI-000000?logo=x&logoColor=white\"){kind=icon}
![\"Node.js](\"https://img.shields.io/badge/Node.js-22%2B-339933?logo=nodedotjs&logoColor=white\"){kind=icon}
![\"MCP](\"https://img.shields.io/badge/MCP-Protocol-8B5CF6\"){kind=icon}
![\"SQLite\"](\"https://img.shields.io/badge/SQLite-Database-003B57?logo=sqlite&logoColor=white\"){kind=icon}
\n synabun.ai · Docs · Blog · Compare · Changelog\n
\n\n---\n\nAny Claude Code, Codex, Gemini, OpenCode, Cursor, Windsurf, or MCP-compatible AI tool can connect to SynaBun and retain knowledge across sessions through [semantic vector search](https://synabun.ai/blog/all-minilm-l6-v2-mcp-vector-memory). Dedicated sidepanels ship for Claude Code, Codex, and OpenCode with tool activity docks, per-agent abort, permission cards, context gauges, and Multi-CLI Resume. Memories are stored in a [local SQLite database](https://synabun.ai/blog/self-hosted-ai-memory-vs-cloud) with built-in vector search — no external services, no API keys, no Docker.\n\n## Features\n\n- **Semantic Search** — find memories by meaning, not keywords, using cosine similarity\n- **Recall Profile Presets** — Quick, Balanced, Deep, and Custom profiles with 6 fine-tune controls, an impact indicator (tokens/recall, memories reachable, session context), and configurable server-side defaults\n- **Multi-Project** — single collection serves all projects with automatic project detection and cross-project recall\n- **Smart Relevance** — time decay (90-day half-life), project boost (1.2x), and access frequency scoring\n- **User-Defined Categories** — hierarchical categories with prescriptive routing descriptions; dynamic schema refresh without server restart\n- **Neural Interface** — interactive 3D force-directed graph visualization at `localhost:3344`\n- **[Local Embeddings](https://synabun.ai/blog/all-minilm-l6-v2-mcp-vector-memory)** — built-in Transformers.js (`all-MiniLM-L6-v2`, 384 dims) — no API key or internet needed after first run\n- **Onboarding Wizard** — guided setup through the Neural Interface UI\n- **Trash & Restore** — soft-delete with full restore capability; trash management via Neural Interface\n- **Memory Sync** — SHA-256 file hash tracking detects when related source files change, flagging stale memories\n- **OpenClaw Bridge** — read-only integration overlays OpenClaw workspace memories as ephemeral nodes in the 3D graph\n- **Category Logos** — upload custom logos for parent categories, rendered on sun nodes in the 3D visualization\n- **Backup & Restore** — export and import SQLite database snapshots via the Neural Interface\n- **[Multi-CLI Sidepanels](https://synabun.ai/blog/skinning-claude-code)** — dedicated sidepanels for Claude Code, Codex, and OpenCode with per-agent abort, tool activity docks, permission cards, TODO drawer, context gauge, session menu, new-session modal with project + branch selection, and Multi-CLI Resume with provider picker\n- **Automation Studio** — visual template editor with user-created collapsible folders (drag/drop, icons, colors), per-template launch defaults (CLI / Model / Thinking effort / MCP profile), and first-class 4-CLI support (Claude Code, Codex, Gemini, OpenCode)\n- **Universal MCP Management** — smart-paste add form auto-detects 5 config formats, syncs one MCP server across Claude Code / Codex / Gemini / OpenCode, centralized env vault with sensitive-value masking, auto profile registration, and one-click install from GitHub\n- **Claude Code Hooks** — 7 lifecycle hooks automate memory recall, storage, and session indexing\n- **User Learning** — 5 binding directives that observe user communication patterns, preferences, and behavioral singularity across sessions (Priority 7 nudge, feature flag, threshold controls)\n- **Hybrid Resume Search** — FTS5 + semantic search across saved sessions so old conversations can be resumed across CLIs\n- **Multi-Runtime Skills** — `/synabun` command hub and `/leonardo` media prompter work on both Claude Code and Codex; `/synabun changelog` quick action available from the OpenCode sidepanel\n- **Leonardo AI Integration** — 5 browser-based MCP tools for AI image and video generation via Leonardo.ai — no API key needed, full UI automation via Playwright with the `/leonardo` skill for guided creation\n- **Enhanced Browser Tooling** — new `browser_cheatsheet` tool with curated selectors per platform, snapshot modes + viewport filtering, `returnSnapshot` that folds actions and observations into one call, selector auto-heal via `textHint`, and CLI/sidepanel session isolation\n- **YouTube Trailer Autopilot** — end-to-end game trailer pipeline: discover candidates via IGDB + Steam, download MP4s (Steam direct or yt-dlp), and publish to YouTube Studio with metadata, thumbnails, scheduling, and analytics — all via browser automation (bypasses the Data API's unaudited-project private-lock and ~6/day quota)\n- **Modern Code Rendering** — syntax highlighting, file path linkification, language labels, hover-only copy buttons, 30fps throttled streaming, and real-time thinking block streaming across every sidepanel\n\n## Table of Contents\n\n- [Quick Start](#quick-start)\n- [Architecture](#architecture)\n- [MCP Tools](#mcp-tools)\n- [Neural Interface](#neural-interface)\n- [Usage](#usage)\n- [Memory Categories](#memory-categories)\n- [Importance Scale](#importance-scale)\n- [How It Works](#how-it-works)\n- [Configuration](#configuration)\n- [Claude Code Hooks](#claude-code-hooks)\n- [Claude Code Skills](#claude-code-skills)\n- [CLAUDE.md Integration](#claudemd-integration)\n- [Multi-Project Support](#multi-project-support)\n- [Known Limitations](#known-limitations)\n- [Troubleshooting](#troubleshooting)\n- [Cost](#cost)\n- [File Structure](#file-structure)\n- [License](#license)\n\n## Documentation\n\n| Document | Description |\n|----------|-------------|\n| **[Website Docs](https://synabun.ai/docs)** | Full documentation on synabun.ai |\n| **[README](./README.md)** | Quick start, architecture, configuration |\n| **[Contributing](./CONTRIBUTING.md)** | Bug reports, feature requests, and forking guide |\n| **[Security](./SECURITY.md)** | Security model and vulnerability reporting |\n| **[Changelog](./CHANGELOG.md)** | Version history |\n| **[Blog](https://synabun.ai/blog)** | Development blog — war stories, architecture decisions, benchmarks |\n| **[Compare](https://synabun.ai/compare)** | Feature comparison with other AI memory tools |\n\n**Contributions:** We welcome [bug reports and feature requests](https://github.com/danilokhury/Synabun/issues). Pull requests are not accepted. See [CONTRIBUTING.md](./CONTRIBUTING.md) for details.\n\n## Quick Start\n\n### Prerequisites\n\n- Node.js 22.19+ (required by the current runtime dependencies)\n\nNo Docker, no API keys, no external services required.\n\n