Everything Setup

Every current Lodis setup path in one place, no quiz required.

Local MCP Clients

Use this when Lodis should run locally through your AI tool. Codex uses TOML; most other MCP clients use JSON.

Fastest path: let Codex write the MCP entry for you.

codex mcp add lodis -- npx -y @sunriselabs/lodis

~/.codex/config.toml

[mcp_servers.lodis]
command = "npx"
args = ["-y", "@sunriselabs/lodis"]

Codex shares this config between the CLI and IDE extension. Use /mcp in the TUI or codex mcp list to confirm Lodis is active.

Agent Instructions

Add the Lodis memory policy to the instruction surface your client reads. Use the general snippet for most clients; Claude Code needs the stronger override shown below.

Use Lodis MCP tools for all persistent memory. At the start of
conversations, call memory_search with relevant terms to retrieve
context. When the user states a preference, corrects an assumption,
shares personal context, or provides information useful across future
conversations, save it with memory_write.

Codex~/.codex/AGENTS.md or project AGENTS.md

Codex AGENTS.md snippet

## Memory — Lodis

Use Lodis MCP tools for persistent memory. Lodis is the source of truth
for preferences, corrections, personal context, project context, and any
fact the user asks you to remember.

### At conversation start
- Call `memory_search` with terms relevant to the user's request to
  retrieve prior context.

### When to write
- `memory_write` — user states a preference, corrects an assumption,
  shares personal context, or says anything useful for future conversations.
- `memory_confirm` — user validates a recalled fact.
- `memory_correct` — user corrects a stored memory.
- `memory_flag_mistake` — a memory turns out to be wrong.

### Rules
- Do not duplicate long-term memories into another memory system when
  Lodis is available.
- Treat Lodis memories as the persistent record across Codex, Claude Code,
  Cursor, Windsurf, and other MCP-connected tools.
- When the user says "remember this," save immediately via `memory_write`.

Claude Code~/.claude/CLAUDE.md or project CLAUDE.md

Claude Code requires a stronger override because it can also read its own file-based memory system.

Claude Code CLAUDE.md snippet

## Memory — CRITICAL OVERRIDE

**DO NOT use the built-in file-based auto-memory system.** Never create,
read, or write to `MEMORY.md` or any files in the
`~/.claude/projects/.../memory/` directory. That system is fully
replaced by Lodis.

**USE Lodis MCP tools for ALL persistent memory.** Lodis is the
single source of truth.

### At conversation start
- Call `memory_search` with terms relevant to the user's request to
  retrieve prior context

### When to write
- `memory_write` — user states a preference, corrects an assumption,
  shares personal context, or says anything useful for future conversations
- `memory_confirm` — user validates a recalled fact
- `memory_correct` — user corrects a stored memory
- `memory_flag_mistake` — a memory turns out to be wrong

### When to search
- `memory_search` — before answering questions where prior context
  would help
- `memory_context` — token-budget-aware retrieval for building responses
- `memory_briefing` — entity summaries (people, projects, etc.)

### Rules
- Never duplicate memories to both Lodis and the built-in file system
- Treat Lodis memories as the persistent record — they survive across
  all MCP-connected tools (Claude Code, Cursor, Windsurf, Claude Desktop)
- When the user says "remember this," save immediately via `memory_write`

Claude DesktopSystem prompt in Settings

Cursor.cursorrules or Rules settings

WindsurfSystem prompt in Settings

First Run

Once Lodis is connected, use the prompt that matches your starting point.

Start fresh"Help me set up Lodis"

Claude memories"Import my Claude memories into Lodis"

ChatGPT memory export"Import this ChatGPT memory export into Lodis"

Cursor rules"Import my .cursorrules as Lodis preferences"

Git config"Import my gitconfig into Lodis"