anchored

README

Anchored

Persistent cross-tool memory for AI coding agents. Single binary. Zero dependencies.

[]

Anchored is an MCP memory server that gives every AI coding agent and IDE you use a shared, persistent memory. Install once, and Claude Code, Cursor, OpenCode, Gemini CLI, and any other MCP-compatible tool read, write, and search the same knowledge base.

No API keys. No daemon. All embeddings run locally.

Features

• Cross-tool memory — one knowledge base, every AI tool and IDE shares it
• Multilingual embeddings — paraphrase-multilingual-MiniLM-L12-v2 (50+ languages, PT-BR and EN parity)
• Hybrid search — RRF fusion of vector similarity (384-dim ONNX) and BM25 (FTS5), with entity and project boost
• Knowledge graph — automatic pattern-based extraction of entities and relationships (no LLM needed)
• Smart categorization — bilingual PT+EN regex auto-classifies memories into 7 categories (fact, preference, decision, event, learning, plan, summary)
• Memory stack — L0 identity + L1 essential stories + L2 on-demand retrieval, budget-enforced (~900 tokens)
• Sandbox tools — anchored_execute/anchored_execute_file run code in 11 languages with stdout-only capture, hardened env, FILE_PATH/FILE_CONTENT injection
• Knowledge indexing — anchored_fetch_and_index mirrors URLs to a local FTS5 store; sandbox keeps raw data out of context
• Background auto-updater — checks GitHub releases on startup; new binary atomically replaced, current process unaffected
• Credential redaction — regex-based secret sanitization before storage
• Multi-source import — Claude Code (JSONL), OpenCode (SQLite), Cursor (.mdc rules), DevClaw

Install

From GitHub Releases:

# Linux / macOS
curl -fsSL https://raw.githubusercontent.com/jholhewres/anchored/main/install/install.sh | bash

From source:

git clone https://github.com/jholhewres/anchored.git
cd anchored && make build
sudo cp bin/anchored /usr/local/bin/

First run auto-downloads the embedding model (~470MB) and creates ~/.anchored/.

Setup

Claude Code (plugin)

The fastest path. Installs the MCP server, six /anchored:* slash commands, and an auto-trigger skill in one step:

/plugin marketplace add jholhewres/anchored
/plugin install anchored@anchored

Then restart Claude Code. From any project: /anchored:context, /anchored:search <query>, /anchored:save <content>, /anchored:stats, /anchored:doctor, /anchored:purge. The skill triggers anchored_* tools proactively when memory is relevant — no need to ask.

Claude Code (MCP only, no slash commands)

claude mcp add -s user anchored anchored

The -s user flag registers Anchored at user scope so it's available in every project. Without it, claude mcp add defaults to local. Entry lives at ~/.claude.json (not ~/.claude/mcp.json). Restart Claude Code to pick up the new server.

Other tools

Run anchored init to auto-detect and register, or configure manually:

Tool	Config file
Cursor	~/.cursor/mcp.json
OpenCode	~/.config/opencode/opencode.json
Gemini CLI	~/.gemini/settings.json
VS Code Copilot	.vscode/mcp.json

{
  "mcpServers": {
    "anchored": {
      "command": "anchored"
    }
  }
}

CLI

anchored                    Start MCP server (STDIO, default when no arg)
anchored serve              Start MCP server (STDIO)
anchored init [--tool]      Auto-detect tools and register MCP
anchored doctor             Diagnose installation, config, MCP registration
anchored stats              Show memory statistics

anchored search <query>     Search memories
anchored save <content>     Save a memory (auto-categorized if --category omitted)
anchored update <id>        Update a memory
anchored forget <id>        Remove a memory (soft delete; --hard for permanent)
anchored list               List memories

anchored import [sources]   Import memories from detected sources
anchored identity [edit]    View or edit identity file
anchored config [show|set]  View or modify configuration

anchored dream              Analyze and consolidate duplicate memories
anchored precompact         Pre-compact memory context
anchored hook <subcommand>  Run session continuity hooks
anchored purge              Wipe memories (--hard for full DB reset with backup)

Import sources: claude-code devclaw opencode cursor all

MCP Tools

Memory

Tool	When to use
anchored_context	First call of every conversation — loads identity, project, recent decisions
anchored_search	Before answering domain questions (hybrid vector + BM25)
anchored_save	Capture facts, preferences, decisions, learnings (category required)
anchored_update	Revise an existing memory in place
anchored_forget	Remove a memory (soft delete by default)
anchored_list	List memories by category, project, or time
anchored_stats	Memory overview
anchored_session_end	Close a tracked session
kg_query	Query knowledge-graph relationships for an entity
kg_add	Capture a relationship (subject — predicate — object)

Sandbox / index (context-saving tools that keep raw data out of context)

Tool	When to use
anchored_execute	Run code in 11 languages; only stdout enters context
anchored_execute_file	Process a file; FILE_PATH and FILE_CONTENT auto-injected
anchored_batch_execute	Run multiple commands and run search queries in one call
anchored_index	Index documentation into FTS5 knowledge base
anchored_ctx_search	Search indexed content with batched queries
anchored_fetch_and_index	Fetch URL → markdown → index (force=true bypasses 24h cache)

How it works

• Hybrid search — RRF fusion of vector similarity (ONNX, multilingual) and BM25 (FTS5), with entity boost and project boost
• Entity detection — extracts project names, tools, and topics from queries to boost relevant results
• Topic change detection — identifies conversation shifts and increases retrieval diversity
• Memory stack — L0 identity + L1 essential stories + L2 on-demand, budget-enforced
• Knowledge graph — bitemporal triples with functional predicates and alias resolution, auto-extracted from memory text
• Credential redaction — regex-based secret sanitization before storage

Storage

~/.anchored/
├── data/
│   ├── anchored.db        # SQLite (FTS5 + vector cache + knowledge graph)
│   └── onnx/              # local embedding model (~470MB)
└── config.yaml

No daemon. No ports. The binary runs on demand via MCP STDIO.

Docs

• Design — memory stack, hybrid search, knowledge graph, quantization
• Architecture — project structure and implementation details
• Embedding Model — model choice, quantization, inference pipeline
• Import Sources — how each tool's data is parsed
• Changelog — version history

License

MIT