Install Lore

Install Lore with the hosted install script:

curl -fsSL https://withlore.ai/install | bash

Then launch Lore with your detected coding agent:

lore run

Lore auto-detects Claude Code, OpenCode, Pi, Codex, and Hermes Agent when you run lore run. For harness-specific setup, see the Guides section.

If you’d rather configure Codex manually (for the Codex Desktop app, or to run codex without going through lore run), run lore setup codex once — it writes ~/.codex/config.toml with the gateway URL and the no-auto-compact override. See the Setup command page for the full reference.

You can also run the gateway directly with npm:

npx @loreai/gateway

Slimmer installs (remote embeddings)

Lore’s on-device (local) embeddings run through @huggingface/transformers and the ONNX runtime — about 480 MB of ML runtime that’s pulled in when you npm install a Lore package (@loreai/core, @loreai/opencode, @loreai/pi). It’s an optional dependency, so if you use a remote embedding provider — or don’t need vector recall — you can skip it:

npm install @loreai/opencode --omit=optional

With the stack absent, recall degrades gracefully to FTS-only keyword search. To keep semantic recall without the local runtime, set a remote provider in .lore.json (search.embeddings.provider = voyage or openai, with the matching API key). The hosted install script and the standalone binary are unaffected — they ship their own runtime and never read node_modules.

Existing Conversations

Lore can import previous coding conversations so a new project memory does not start from a blank slate:

lore import

Imported history feeds the same distillation and knowledge pipeline as live sessions. lore import detects Claude Code, Codex, OpenCode, Pi, Aider, Cline, and Continue, lets you pick which agents to import, and finds conversations across all of a repo’s git worktrees.

→ See the full Import conversations guide.