# max-ramas/rms-memory-mcp [Health: Active]

**Category:** 🧠 Knowledge & Memory  
**Repository:** https://github.com/max-ramas/rms-memory-mcp  
**GitHub Stars:** 38  
**Views:** 3  
**Installs:** 0  
**Upvotes:** 0  
**Directory Page:** https://allmcps.com/mcp/max-ramas-rms-memory-mcp

## Description
Local-first persistent memory for AI coding agents. Hybrid vector + full-text search (LanceDB + Tantivy), separate Markdown and Tree-sitter code indexes, per-project isolated vaults shared across IDEs (Cursor, Zed, Claude Code, Codex), bounded recall with abstain-on-low-confidence.

## Claude Desktop Quick Installation
Remote MCP endpoint (confidence: high). Install path detected from listing signals. Add as a URL/SSE server in your client:

```json
"mcpServers": {
  "rms-memory-mcp": {
    "url": "https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT"
  }
}
```

## Documentation & README

<div align="center">

# 🧠 RMS Memory MCP

**Version:** `1.1.0` (2026-07-27) · companion GUI `1.1.0` (unified numbering)

**Persistent, local-first memory for your AI coding agents.**

Stop re-explaining your architecture to Cursor, Zed, and Claude Code and other IDEs every single session.

[![License: MIT](https://img.shields.io/badge/License-MIT-yellow.svg)](https://opensource.org/licenses/MIT)
[![Rust](https://img.shields.io/badge/Rust-1.75%2B-orange?logo=rust)](https://www.rust-lang.org/)
[![Crates.io](https://img.shields.io/crates/v/rms-memory-mcp)](https://crates.io/crates/rms-memory-mcp)
[![Release](https://img.shields.io/github/v/release/max-ramas/rms-memory-mcp?color=blue)](https://github.com/max-ramas/rms-memory-mcp/releases)
![Downloads](https://img.shields.io/github/downloads/max-ramas/rms-memory-mcp/total)
![Build](https://github.com/max-ramas/rms-memory-mcp/actions/workflows/release.yml/badge.svg) 
[![Platform](https://img.shields.io/badge/platform-macOS%20%7C%20Linux%20%7C%20Windows-lightgrey)]()
[![MCP](https://img.shields.io/badge/protocol-MCP-blueviolet)](https://modelcontextprotocol.io)

[Features](#-key-features) • [Download](https://github.com/max-ramas/rms-memory-mcp/releases/latest) • [Install](#-installation) • [Quick Start](#-quick-start) • [CLI](#-cli-commands) • [MCP Tools](#-mcp-tools-exposed) • [Architecture](#-architecture-highlights)

</div>

---

<div align="center">
  <a href="https://ko-fi.com/M7I020HKXX" target="_blank" rel="noopener noreferrer">
    <img src="https://ko-fi.com/img/githubbutton_sm.svg" alt="ko-fi">
  </a>
</div>

---

## The Problem

You're developing a single project but switching between different agents — Cursor, Zed, Claude Code, OpenCode, etc. Every one of them loses context of architectural decisions, system requirements, and user preferences the moment you close the tab. You end up re-explaining the same things over and over, or copy-pasting a stale `CLAUDE.md` between tools.

**RMS Memory MCP** bridges this gap: a single, isolated, centralized Markdown vault — perfectly structured for LLM consumption — that any MCP-compatible IDE can read from and write to.

## ✨ Key Features

| | |
|---|---|
| 🗂️ **Global Centralized Vaults** | Project context lives outside your repo — zero `.mcp` file pollution. |
| 🔍 **Hybrid Retrieval (LanceDB)** | Embedded Vector Search + Tantivy Full-Text Search for zero-fail context hits. |
| 🌐 **Multilingual Semantic Parsing** | `fastembed-rs` + `multilingual-e5-small` — native Russian & English understanding. |
| 🌳 **AST Markdown Chunker** | `pulldown-cmark`-based chunking keeps code blocks and lists bound to their parent heading. |
| 🧩 **Semantic Code Memory** | Optional Tree-sitter indexing for Rust, Go, JS/JSX, TS/TSX, Python, C/C++, Java, Ruby, Swift, and Vue `<script>` blocks; stable segment identities and repeated preambles preserve context when large implementations split. Opt-in `watch` mode reindexes **only dirty paths** (with full-walk fallback). |
| 🕸️ **Knowledge Graph Foundation** | Derived Markdown/code relationships and durable user overrides are stored separately from retrieval chunks and consumed by the companion GUI graph. |
| 🧹 **Safe Project Lifecycle** | Unregistering preserves vault/index data; permanent GUI deletion requires the exact project key, is confined to the master vault, and never touches source code. |
| 🔀 **Federated Corpus Search** | Search `vault`, `code`, or `all`; mixed results use Reciprocal Rank Fusion rather than incompatible raw vector distances. |
| 🎯 **Bounded Recall (v1.0.7)** | `rms_search` returns an inject/abstain envelope with `max_chars`, optional `min_score`, fail-closed errors, and `retrieval_mode` (`hybrid` or short-query `fts_prefer`). |
| ♻️ **Knowledge Lifecycle** | Frontmatter `status` / `supersedes` / temporal `valid_*` gate Lance recall; soft supersede via `rms_write`; Doctor freshness lint (7/7). |
| 🔄 **Session Continuity (v1.0.7)** | Vault-backed checkpoints (`rms_checkpoint_save/done/load/query`), `rms_overview` project orientation, `rms_system_instructions` self-bootstrap, editor-agnostic `rms-memory hook` CLI, installer L3 thin adapters, and `pinned` notes that bypass temporal/`min_confidence` recall gates. |
| 🧭 **Multi-project MCP routing (v1.0.8+)** | Explicit `project` always rebinds the active vault; empty Cursor `roots/list` falls back to process cwd; injected rules require `project: "<key>"` on every memory tool call; `rms-memory inject-rules [--all]` refreshes keys. |
| 🔗 **Cross-project federated search (v1.0.9)** | Pass `projects: [key, …]` to `rms_search` / `rms_code_search` for read-only RRF federation. Vault/all across multiple projects requires `cross_project_vault=true` on every listed key (hard fail otherwise). |
| 🧊 **Concurrent bind cache (v1.0.9)** | Up to 4 warm Store+watcher pairs (LRU); multi-root IDE sessions stop thrashing open/close. |
| 🧱 **Cargo workspace (v1.0.9)** | `rms-memory-{core,index,vault}` path crates under `crates/`; public umbrella `rms-memory-mcp` keeps stable module paths for the GUI. See [docs/crate-split.md](./docs/crate-split.md). |
| 📦 **Unified Releases** | Public assets use `rms_memory_mcp_<version>_<target>.*` / `rms_memory_gui_<version>_*` on the **same** `vX.Y.Z` tag (MCP + GUI share numbering). Unversioned names are no longer published. |
| ⚙️ **Dynamic Auto-Installer** | `rms-memory install` scans your system and wires itself into every supported IDE. |
| 📜 **Rules-as-Code Patching** | Non-destructive AST patching of `.cursorrules`, `.zed/assistant.md`, etc. Opt-in by default. |
| 🧪 **Durable Vault Writes** | `rms_write` creates rolling `.bak` backups and atomically replaces `create`/`replace` targets after fsync, so interrupted writes never expose a truncated Markdown file. |
| 📚 **Canonical Wiki Isolation** | Generated `<vault>/wiki/**` stay Git-synchronized but are excluded from indexes/search/watchers/graph/packs; MCP write and canonical DocumentService also reject wiki paths (wiki-safe writers only). |
| 🛡️ **Ten-Point Resiliency** | GC, background sync, write-guard snapshots, macOS sandbox bypass, `llms.txt` export, path traversal + injection protection, zombie prevention, graceful shutdown. |
| 🔒 **Security Hardened** | Panic-free database layer, symlink traversal blocked, JSON-RPC error responses, request size limits. See [SECURITY.md](./SECURITY.md) and [NOTICE](./NOTICE). |
| 🧠 **Audit Metadata** | Every record auto-receives `last_modified_by`, `timestamp`, `confidence`, `source` — agents can filter by reliability. |
| 🔀 **Multi-Scope** | `--scope` flag supports arbitrary identifiers beyond filesystem paths (thread IDs, lead IDs, etc.). |
| 🖥️ **Optional Companion GUI** | Paid Tauri desktop app: visual Markdown/graph editor, Git & Vault sync, Doctor dashboard, AI-assisted organizer/Wiki (BYOK), and cross-tool spend tracking — layered on top of the same vault, never required. See [GUI-README.md](./GUI-README.md). |

## 📦 Installation

### Option 1: Homebrew (macOS Apple Silicon & Linux)

```bash
brew tap max-ramas/tap
brew install rms-memory-mcp
```

Installs a prebuilt binary — no Rust toolchain required. The formula updates
automatically with every release.

> **Not covered by Homebrew:** macOS Intel (dropped as of v1.0.1) and
> Windows (Homebrew doesn't run there — use Option 2 or the `.zip` below).

### Option 2: GitHub release binary

Prebuilt binaries for `aarch64-apple-darwin` (Apple Silicon), `x86_64-unknown-linux-gnu`,
`aarch64-unknown-linux-gnu`, and `x86_64-pc-windows-msvc` are published on every
[release](https://github.com/max-ramas/rms-memory-mcp/releases), along with
`.deb`/`.rpm` packages for Linux. One-line installers auto-detect your architecture:

```bash
curl -fsSL https://raw.githubusercontent.com/max-ramas/rms-memory-mcp/master/scripts/install.sh | bash
```

```powershell
irm https://raw.githubusercontent.com/max-ramas/rms-memory-mcp/master/scripts/install.ps1 | iex
```

### Option 3: Build from Source

```bash
# 1. Clone the repository
git clone https://github.com/max-ramas/rms-memory-mcp.git
cd rms-memory-mcp

# 2. Build the optimized release binary
cargo build --release

# 3. Add the binary to your global PATH
cp target/release/rms-memory ~/.cargo/bin/
```

### crates.io (`cargo install`) — lagging until workspace publish

As of **1.0.9** the library is a Cargo workspace with path-only internal crates
(`publish = false`). `cargo publish` of the umbrella is intentionally disabled.
Prefer Homebrew or a GitHub release binary for the current version.

```bash
# May install an older crates.io revision until the publish strategy lands
# (see docs/crate-split.md). Not the recommended path for 1.0.9+.
cargo install rms-memory-mcp
```

### Optional RMS Memory GUI installers

The companion **RMS Memory GUI** is a paid, optional Tauri desktop control plane:
a visual Markdown/graph editor, per-project and Vault-wide Git/GitHub sync, a
Doctor dashboard with one-click repair, an AI-assisted organizer and Wiki
generator (bring your own key, proposal-only), and cross-tool spend tracking.
The MCP server remains fully standalone: it does not require the GUI, an AI provider,
or a GUI license to index, search, sync or serve MCP clients.

See [GUI-README.md](./GUI-README.md) for the full feature breakdown, supported
platforms, installer verification and release-distribution policy.

GUI source is private, but desktop installers are published as binary assets
on this repository's [GitHub Releases](https://github.com/max-ramas/rms-memory-mcp/releases)
under the matching `v<version>` tag. Until Apple/Windows signing certificates
exist, macOS builds may be unsigned — see [GUI-README.md](./GUI-README.md) for
Gatekeeper notes. The private GUI workflow transfers only the completed
`.dmg`, `.msi`/`.exe`, `.AppImage`, `.deb`, and `.rpm` installer files (plus
`SHA256SUMS.txt` when present); it never mirrors GUI source, build logs,
updater metadata, credentials, or other release archives. The same publication
flow runs for a `v*` GUI tag and for a manually dispatched, version-validated
GUI release.

## 🚀 Quick Start

The fastest way to get every IDE on your machine connected:

```bash
rms-memory install
```

This scans `~/.config/` and `~/Library/Application Support/` and hooks `rms-memory` directly into **Cursor**, **Zed**, **Claude Code**, **OpenCode**, and others — no manual JSON editing.

### Generated Wiki namespace

The optional desktop GUI writes human-readable Wiki pages to `<vault>/wiki/`. RMS Memory MCP remains AI-free and treats this directory as generated output rather than canonical memory. A shared case-insensitive path policy (`src/path_policy.rs`, also reused by the GUI) excludes the entire namespace from Markdown/code indexing, vector and full-text retrieval, watchers, the durable graph and Wiki context packs. **Write isolation** matches that policy: `rms_write` requires `.md` and rejects `wiki/**`; canonical `DocumentService` list/read/write APIs exclude or reject wiki; Wiki page mutations use wiki-safe methods that skip memory audit-frontmatter injection. Linked-document `link:` resolution always re-checks that the canonical target stays inside the vault. Full or incremental sync removes legacy Wiki-derived records by path without deleting the files, and `doctor` reports the isolation state explicitly.

For virtual projects without a filesystem path (threads, leads, etc.), use `--scope`:

```bash
rms-memory --scope "thread:abc-123" serve
```

### Use multiple isolated scopes

A scope is an isolation boundary for a vault and its index. Without `--scope`, RMS Memory uses the canonical current working directory; an explicit filesystem path addresses that same kind of project vault. Any other non-empty identifier creates an isolated virtual vault:

```bash
rms-memory serve                                      # current project scope
rms-memory --scope "/home/user/my-project" serve     # explicit project scope
rms-memory --scope "thread:abc-123" serve             # virtual thread scope
rms-memory --scope "product:acme" serve                # virtual product scope
```

For project knowledge plus per-thread history, query each scope explicitly and merge the results in the caller. RMS Memory intentionally does not mix scopes implicitly. Scope IDs may not be empty or exceed 512 characters; absolute and `./`/`../` values are resolved as paths, while all other values are opaque identifiers.

When using `min_confidence`, start with an unfiltered search. Use `0.3–0.5` for broad refinement and reserve `0.7+` for verified canonical facts; records without a confidence value remain visible.

### Configure your vault

The simplest way to configure the server is to run the interactive setup wizard. You don't need to memorize any CLI flags — just run:

```bash
rms-memory config
```

*(Alternatively, set the vault root directly with `rms-memory config --vault-path ~/MyVaults/`, then run `rms-memory init` in each repository you want to register.)*

Register a repository explicitly from its root before connecting IDE agents:

```bash
cd /path/to/project
rms-memory init
```

This creates the project mapping in `~/.rms-memory/registry.toml` and provisions its isolated, structured vault. Routine MCP discovery is read-only and fail-closed: it never creates a project from `/`, never falls back to a shared global vault, and never guesses between multiple registered projects.

```text
~/MyVaults/
  └── <ProjectKey>/
      ├── rules/
      ├── decisions/
      ├── architecture/
      ├── artifacts/
      ├── docs/
      └── api/
```

### Optional semantic code memory

Markdown memory remains the default corpus. Semantic source indexing is separate, supports all bundled language adapters, and never changes source files:

```bash
rms-memory reindex --code  # build/update only derived code memory
rms-memory reindex --all   # refresh Markdown vault + code memory
```

Registered projects support `code_index_mode = "off" | "manual" | "watch"`; the default is `off`. Set it from the project root with `rms-memory config --code-index-mode watch` (or add `--scope <project-path>`). `watch` is explicitly opt-in, coalesces supported source saves for three seconds, and **reindexes only the dirty paths** (`try_index_code_paths`) with a full-walk fallback when the index is cold, the dirty set is empty/oversized (>200), or the watcher channel overflows. Concurrent IDE processes share a completion marker so an unchanged workspace stays idle. Code search results include their source language.

Perf smoke for large fixtures: `./scripts/bench_large_vault.sh [notes] [code_files]`.

Language selection is project-scoped and defaults to every bundled adapter:

```bash
rms-memory config --code-languages auto
rms-memory config --code-languages go,typescript,tsx,vue
```

Supported names are `rust`, `go`, `javascript`, `jsx`, `typescript`, `tsx`, `python`, `c`, `cpp`, `java`, `ruby`, `swift`, and `vue`. Generated paths (`node_modules`, `.next`, `.nuxt`, `target`, `vendor`, and `coverage`) are always excluded. Ambiguous `.h` files are indexed as C exactly once; use `.hpp`, `.hh`, or `.hxx` for C++ headers. Vue indexes only inline JavaScript/TypeScript `<script>` contents and maps results back to the `.vue` host file; templates, styles, `script setup` macros, and external `src` scripts remain outside v1.0.5 semantic extraction.

## 🛠 CLI Commands

| Command | Description |
|---|---|
| `rms-memory serve` | Starts the JSON-RPC stdio server (auto-triggered by your IDE). |
| `rms-memory init` | Registers a project into the global registry. `--dry-run` supported. `--full` forces creation of all IDE rule templates. |
| `rms-memory inject-rules [--all]` | Re-injects managed IDE rule blocks with the concrete registry `project` key (existing files only, unless `--full`). Use after template updates. Fail-closed: a single unregistered path is refused rather than injected with a guessed key — run `init` there first, or use `--all` to refresh every registered project. |
| `rms-memory import` | Scans for existing docs (`README.md`, `docs/`, `ADR/`) and imports them — interactively or via `--auto-import`. |
| `rms-memory install` | Hooks the server into supported IDEs. `--dry-run` supported. |
| `rms-memory uninstall` | Removes the server from all discovered IDE configurations. |
| `rms-memory doctor` | Runs 5-point vault health diagnostics. `--repair-frontmatter` safely repairs duplicate, missing, and known attached frontmatter IDs with backups; arbitrary invalid YAML is reported but never rewritten automatically. |
| `rms-memory config` | Without flags: prints global + current-project settings, then offers interactive global editing. Any flag runs non-interactively. Global: `--vault-path`, `--auto-add`, `--inject-rules`, `--auto-import skip\|link\|import_organize\|import`, `--max-backups N`. Project (cwd or `--scope <path>`): `--code-index-mode off\|manual\|watch`, `--code-languages auto\|<comma-list>`, `--include <globs>`, `--exclude <globs>`, `--cross-project-vault true\|false`. |
| `rms-memory reindex [--vault\|--code\|--all]` | Refreshes Markdown memory (default), derived semantic code memory, or both. |
| `rms-memory sync` | Incremental LanceDB delete-then-insert sync (also runs automatically during `serve`). |
| `rms-memory gc` | Prunes orphaned LanceDB indices belonging to deleted vaults. |
| `rms-memory log` | Tails the telemetry log (`~/.rms-memory/rms.log`). |
| `rms-memory export-llms` | Compiles the current vault into a single `llms.txt` payload. |
| `rms-memory projects list` | Lists registered project keys and their code/vault paths. |
| `rms-memory projects locate --project <key>` | Resolves one registered project key. |
| `rms-memory projects remove <key>` | Removes an erroneous project registration while preserving its vault files. |
| `rms-memory hook --event <e>` | Editor-agnostic continuity hook (`session_start`, `pre_compact`, `session_stop`); JSON on stdout. `--project <key>` or unique cwd resolution (fail-closed); `--apply` creates/updates or closes a checkpoint. |
| **All commands** | Accept `--scope <id>` to target arbitrary isolated vaults (threads, leads, etc.). |

## 🔌 MCP Tools Exposed

Tool descriptions are written to be **action-oriented**, so agents use the vault proactively without being asked.

<table>
<tr><th>Tool</th><th>Purpose</th><th>Input</th></tr>
<tr>
<td><code>rms-memory_rms_search</code></td>
<td>Searches Markdown memory by default. Set <code>corpus</code> to <code>code</code> or <code>all</code>; <code>all</code> uses Reciprocal Rank Fusion. Pass <code>projects: [key, …]</code> for read-only cross-project federation (mutually exclusive with <code>project</code>); vault/all with <code>len&gt;1</code> requires <code>cross_project_vault=true</code> on every listed key (hard fail — no silent degrade). Returns an inject/abstain decision envelope (<code>decision</code>, <code>reason</code>, <code>injected_ids</code>, optional <code>retrieval_mode</code>). Agents are instructed to call this <em>first</em>.</td>
<td><code>{ query, project?, projects?, corpus: vault|code|all, limit, include_content, min_confidence, max_chars?, min_score? }</code></td>
</tr>
<tr>
<td><code>rms-memory_rms_code_search</code></td>
<td>Convenience endpoint for the derived semantic code index. Results include file, symbol, kind, line range, and segment index. Same optional <code>projects: […]</code> federation as <code>rms_search</code> (code-only; does not change the active bind).</td>
<td><code>{ query, project?, projects?, limit, include_content }</code></td>
</tr>
<tr>
<td><code>rms-memory_rms_read</code></td>
<td>Reads the full contents of a document found via <code>rms_search</code>.</td>
<td><code>{ path, project? }</code></td>
</tr>
<tr>
<td><code>rms-memory_rms_write</code></td>
<td>Persists new decisions, constraints, or rules. Agents are prompted to call this <em>proactively</em> after solving a tricky bug or learning a preference. Auto-injects audit metadata. Optional soft supersede of a prior note.</td>
<td><code>{ path, project?, content, mode: replace|append|create, confidence, source, status?, supersedes? }</code></td>
</tr>
<tr>
<td><code>rms-memory_rms_projects</code></td>
<td>Lists registered project keys even when the MCP client did not supply workspace roots.</td>
<td><code>{}</code></td>
</tr>
<tr>
<td><code>rms-memory_rms_overview</code></td>
<td>Structured orientation for exactly one project: counts by folder/status, recent notes, active checkpoints, coverage metadata. Fail-closed scoping; never aggregates across projects. Returns <code>structuredContent</code> + JSON text.</td>
<td><code>{ project?, recent_limit? }</code></td>
</tr>
<tr>
<td><code>rms-memory_rms_checkpoint_save</code> / <code>done</code> / <code>load</code> / <code>query</code></td>
<td>Session continuity on plain Markdown (<code>artifacts/checkpoints/</code>): save goal/pending/links before context compaction, list or load active checkpoints on resume, close with a summary — closing writes a durable session note under <code>artifacts/sessions/</code> and drops the checkpoint out of default recall.</td>
<td><code>{ name, project?, goal?, pending?, links?, summary?, status? }</code></td>
</tr>
<tr>
<td><code>rms-memory_rms_system_instructions</code></td>
<td>Returns the canonical memory-usage protocol (search-first, persist, continuity) so agents can self-bootstrap without injected rule files.</td>
<td><code>{ project? }</code></td>
</tr>
</table>

The server resolves an explicit scope or legacy `rootUri`, then negotiates MCP `roots/list`. If a client exposes neither (or opens several registered roots), pass the short registry key in `project`; injected agent rules contain the correct key for that repository. `rms_projects` lists valid keys without requiring a bound workspace. An explicit `project` on any tool call always wins and rebinds the active vault — one long-lived MCP process can serve every registered project. Without `project`, ambiguity stays fail-closed (no silent pick-first).

To remove an accidental registration without deleting its Markdown vault:

```bash
rms-memory projects remove <key>
```

The CLI command is intentionally non-destructive. The companion GUI exposes a
separate **Delete project and data** action for permanent cleanup of the
registration, Markdown vault, and derived index. It requires typing the exact
project key and accepts only a dedicated child of the configured master vault;
the repository source path is explicitly excluded from deletion.

## 🏗 Architecture Highlights

<details>
<summary><b>Cargo workspace (v1.0.9)</b></summary>

Implementation lives in path-only crates under `crates/` (`rms-memory-core`, `rms-memory-index`, `rms-memory-vault`; `rms-memory-cli` reserved). The published product remains the root umbrella `rms-memory-mcp` (binary + MCP server + tools + rules injector), which re-exports every former `rms_memory_mcp::<module>` path so the companion GUI keeps stable imports. Heavy deps (`lancedb`, `ort`, `fastembed`, tree-sitter) concentrate in `rms-memory-index`. Details: [docs/crate-split.md](./docs/crate-split.md).
</details>

<details>
<summary><b>Unified Configuration & Knowledge Isolation</b></summary>

A central `~/.rms-memory/registry.toml` routes every project to an isolated vault, computed from a hash of the project path. No `.mcp` files, no per-repo config — global MCP entries (e.g. Zed's `settings.json`) can target any workspace automatically.
</details>

<details>
<summary><b>Safe Project Lifecycle</b></summary>

The transport-neutral `ProjectService` is the single implementation used by
the CLI and companion GUI. Registry mutation remains revisioned through
`ConfigManager`; deletion validates canonical paths before unregistering and
returns structured warnings if filesystem cleanup cannot be completed.
</details>

<details>
<summary><b>Linked Documents (zero-copy import)</b></summary>

Instead of duplicating existing docs into the vault, `rms-memory import` can create lightweight **Link Files** — Markdown stubs with a `link: <path>` frontmatter property. Reads/writes are transparently redirected to the source file, while the vector index still respects the vault's directory structure.
</details>

<details>
<summary><b>Hybrid Search (LanceDB + Tantivy)</b></summary>

Embedded LanceDB (`~/.rms-memory/dbs/`) combines vector similarity with full-text search, so a query never comes back empty just because the exact keywords didn't match.
</details>

<details>
<summary><b>Separate Markdown and Code Corpora</b></summary>

Human-authored Markdown and derived Rust code live in separate tables. Code chunks carry stable symbol identities, line ranges, and preambles; unchanged chunks reuse their vectors. `corpus=all` fuses independently ranked result sets with Reciprocal Rank Fusion, avoiding any assumption that distances from the two corpora are comparable.
</details>

<details>
<summary><b>Graph-ready Knowledge Core</b></summary>

Graph nodes and edges are deliberately independent of retrieval chunk boundaries. Markdown links, Rust imports, trait implementations, and lexical call hints can be reconciled as derived relationships; user-created edges and suppress/restore overrides persist across reindexing. Current Rust call edges are syntax-level hints, not a compiler-accurate call graph.
</details>

<details>
<summary><b>AST-Aware Chunking</b></summary>

`pulldown-cmark` parses the Markdown AST directly. Chunks are built by walking up to the parent heading, with a strict 1500-character boundary and ~200-character overlapping window for oversized code blocks — no mid-sentence truncation.
</details>

<details>
<summary><b>Ten-Point Production Resiliency</b></summary>

1. Path traversal + filter injection prevention
2. Zombie process prevention (watcher shutdown on EOF + `std::process::exit(0)`)
3. Graceful shutdown (`SIGINT`/`Ctrl+C` handler)
4. macOS sandbox bypass for `fastembed` model downloads
5. `rms-memory gc` — orphaned vector store pruning
6. PID-aware per-project writer lock and read-only background synchronization across IDE processes
7. Markdown watcher plus an explicitly opt-in, 3s-debounced code watcher with **path-scoped** reindex and shared-generation suppression
8. Write-guard snapshotting with rolling `.bak` backups (default: 5)
9. Isolated telemetry logging (`~/.rms-memory/rms.log`)
10. `llms.txt` export for flat, decoupled LLM ingestion
</details>

<details>
<summary><b>Validated v1.0.5 Multi-IDE Behavior</b></summary>

Live MCP requests have been verified for `rms_search(corpus=vault|code|all)` and `rms_code_search`. On this repository, `reindex --code` indexed 43 Rust files into 298 semantic items and 438 segments with all vectors reused on an unchanged run. An isolated five-server watcher run coalesced rapid saves into one shared completion-marker update; a later real-project stress gate completed concurrent GeoMail, License Server, RMS Monitoring, and GeoTax Site indexing, then seven MCP servers (after four IDE restarts) stayed at 0.0% CPU with no background reindex.
</details>

## 🧩 Supported IDEs

| IDE | Auto-Install | Rules Injection |
|---|:---:|:---:|
| Cursor | ✅ | `.cursorrules` |
| Zed | ✅ | `.zed/assistant.md` |
| Claude Code | ✅ | `.claude/CLAUDE.md` |
| OpenCode | ✅ | — |
| Codex | ✅ | — |
| VS Code | ✅ | — |
| Antigravity | ✅ | — |

## 📄 License

MIT License — see [LICENSE](LICENSE) for details.

---

<div align="center">
<sub>Built by <a href="https://ramzaeff.com">Maksim Ramzaev</a> · <a href="https://rms-ds.com">RMS Digital Services</a></sub>
</div>