Add sovseal Memory to Windsurf AI Editor

sovseal adds a persistent, local-first memory layer to Windsurf’s Cascade AI assistant. Once configured, Cascade can store and retrieve design decisions, coding conventions, configuration paths, and developer preferences across project workspaces — no credentials or cloud setup required for local storage.

Installation

Windsurf Settings UI
Config file

Open Windsurf Settings

Press Cmd+, (macOS) or Ctrl+, (Windows/Linux) to open the settings panel.

Navigate to MCP

Scroll to the MCP (Model Context Protocol) section and click Add Server.

Enter the server configuration

Fill in the following fields:

Field	Value
Name	`sovseal-memory`
Type	`stdio`
Command	`npx`
Arguments	`-y @sovseal/mcp-server`

Save and restart

Click Save, then close and reopen Windsurf. The store_memory and recall_memory tools will be available to Cascade in the chat panel.

Edit ~/.codeium/windsurf/mcp.json directly and add the sovseal-memory block:

~/.codeium/windsurf/mcp.json

{
  "mcpServers": {
    "sovseal-memory": {
      "command": "npx",
      "args": ["-y", "@sovseal/mcp-server"]
    }
  }
}

Save the file and restart Windsurf to pick up the new server.

Direct Cascade with custom instructions

To make Cascade proactively search memory at the start of every session, create a .windsurfinstructions file at the root of your project (or .codeium/instructions.txt):

.windsurfinstructions

Always query the `recall_memory` tool first when starting a session or working on
a new feature to retrieve relevant project context, past developer decisions, and
stack preferences.
Call `store_memory` whenever the user establishes a new convention, configuration
path, or coding style preference.

Cascade reads this file before each conversation, so the instruction takes effect automatically without any manual prompting.

Project-scoped memory isolation

By default, sovseal stores memories globally at ~/.sovseal/db/memories.lance. To prevent context from one codebase leaking into another, set SOVSEAL_DB_DIR in your project-level mcp.json:

.codeium/windsurf/mcp.json

{
  "mcpServers": {
    "sovseal-memory": {
      "command": "npx",
      "args": ["-y", "@sovseal/mcp-server"],
      "env": {
        "SOVSEAL_DB_DIR": "./.sovseal/db"
      }
    }
  }
}

With this setting, Cascade only accesses memories stored inside the current project’s folder. Memories from other projects remain completely separate.

On first launch, sovseal downloads the quantized embedding model (~22 MB) to ~/.sovseal/models/. Windsurf may show Cascade as “thinking” for up to 30 seconds during this one-time download. All subsequent sessions start instantly from the local cache.

Get Started

Platform

Agent Plugins

Self-Hosted

Integrations

Cookbooks

API Reference

Release Notes

Add sovseal Memory to Windsurf AI Editor

Installation

Direct Cascade with custom instructions

Project-scoped memory isolation

​Installation

​Direct Cascade with custom instructions

​Project-scoped memory isolation

Installation

Direct Cascade with custom instructions

Project-scoped memory isolation