Skip to content

Quickstart

This page assumes you have already followed the Installation steps and have at least one provider authenticated.

1. Bootstrap the workspace

In any project you want Lime to work on, run:

Terminal window
lime init

lime init writes:

  • An AGENTS.md you can edit to describe project conventions.
  • A .lime/ directory used for per-directory context, sessions, and local settings.
  • A .lime.json file for shared (committable) repo defaults.
  • The relevant .gitignore entries so machine-local state doesn’t leak.

2. Start the REPL

Terminal window
lime

The REPL boots with the workspace’s AGENTS.md, project .lime/ instructions, and any user-level memory loaded automatically. Tokens stream character-by-character, tool calls render inline, reasoning summaries collapse and expand, and Lime prompts before any side effect.

> find every TODO and group them by file
Searching workspace…
✓ glob_search **/*.rs (124 files)
✓ grep_search "TODO" (12 hits in 7 files)
✓ read_file src/auth.rs
## TODOs by file
**src/auth.rs** (3)
- line 42 — TODO: add refresh-token rotation

3. Useful keystrokes

KeyAction
/Open the slash-command palette (fuzzy matched).
!Drop into a one-shot shell line — stdout is persisted as a session message.
Recall previous prompt.
EscCancel the current turn.

4. Pick a model

Terminal window
lime --model claude-opus-4-7
lime --model o4-mini --reasoning-effort high
lime --model pro # alias for gemini-2.5-pro

Or persist your choice in .lime/settings.local.json:

{ "model": "claude-opus-4-7" }

The default model when none is set is gpt-5.4. See Models & providers for the full catalog.

5. Try non-interactive mode

You don’t have to start the REPL — you can run a single slash command non-interactively against the most recent session:

Terminal window
lime --resume latest /status
lime --resume latest /commit

For a full programmatic workflow, see Non-interactive use.

6. Where to next