Multi-Platform Support

Lavra supports four AI coding agents. The core memory system (hooks, knowledge capture, auto-recall) works identically across all platforms. Commands, agents, and skills are available on all platforms.

What works where

Feature	Claude Code	OpenCode	Gemini CLI	Cortex Code
Memory capture	✓	✓	✓	✓
Auto-recall	✓	✓	✓	✓
Commands	✓	✓	✓	✓
Agents	✓	✓	✓	✓
Skills	✓	✓	✓	✓
Context7 MCP	✓	✓	✓	manual

Claude Code

npx @lavralabs/lavra@latest --claude       # local project
npx @lavralabs/lavra@latest --global       # all projects (~/.claude/)

OpenCode

npx @lavralabs/lavra@latest --opencode           # local project
npx @lavralabs/lavra@latest --opencode --yes     # skip model selection prompts

The installer copies a TypeScript plugin to .opencode/plugins/lavra/ (local) or ~/.config/opencode/plugins/lavra/ (global) and installs dependencies with Bun. Commands, agents, and skills are converted to OpenCode format automatically.

You’ll be prompted to choose which models to map to each tier (haiku/sonnet/opus). See Model Selection for details.

Verify:

ls -la .opencode/plugins/lavra/plugin.ts
ls -la .opencode/hooks/

Check plugin is loading (look for these in OpenCode output):

[lavra] Plugin loaded successfully
[lavra] session.created hook triggered

Gemini CLI

npx @lavralabs/lavra@latest --gemini       # local project

The installer converts commands to .toml format and copies commands, agents, skills, and hooks to .gemini/ (local) or ~/.config/gemini/ (global). Memory capture and auto-recall work via the same stdin/stdout JSON protocol as Claude Code. Context7 MCP is configured automatically in ~/.config/gemini/settings.json.

Verify:

ls -la .gemini/hooks/
cat gemini-extension.json | jq '.hooks'

Cortex Code

bash /path/to/lavra/installers/install-cortex.sh

The installer copies hooks to .cortex/hooks/ (local) or ~/.snowflake/cortex/hooks/ (global). Commands, agents, and skills use .md format (same as Claude Code). Hooks are configured via hooks.json.

Context7 MCP is not installed automatically. To enable framework documentation lookup, add it manually to ~/.snowflake/cortex/mcp.json:

{
  "mcpServers": {
    "context7": {
      "url": "https://mcp.context7.com/mcp",
      "type": "http"
    }
  }
}

Cortex Code also reads .claude/ directories for compatibility, but native .cortex/ paths are preferred.