Slash commands

Type / at the start of the AI message box and a list of commands appears. It narrows as you type. Pick one with ↑ ↓ and Enter or Tab, or click it. Esc closes the list. You can also type the whole command, with its argument, and press Enter.

Only these exact names are commands. A message that starts with a path, like /home/me/app is slow, is sent as an ordinary message.

/model

/model on its own opens a two-step picker. First choose a provider: the list shows the ones you can use right now (every provider you've saved a key for, and a local model), plus Configure another provider…. Then choose one of that provider's models, fetched live from the provider.

/model <model-id> sets a model directly on the current provider, for example /model gpt-5. Use this for a model the list doesn't show. The chat confirms the switch, for example Switched to OpenAI · gpt-5. Model ids can't contain spaces.

See Models and providers to add a provider.

/fim

Chooses the model for autocomplete, separately from the chat model. Typing sends a request at every pause, so a small, fast model is usually best here.

• /fim opens a picker: Automatic, then each provider you can use, then its models.
• /fim <model-id> sets a model directly on the provider already used for autocomplete (or the chat provider, if none is chosen).
• /fim auto goes back to automatic. default, reset and off do the same.

Automatic uses Mistral's Codestral when you've saved a Mistral key, since it's built for completing code in the middle of a file. Otherwise it uses your chat provider: its fast model for Gemini, and the selected chat model for the others.

/clear

Stops any running task and empties the conversation. The next message starts with no memory of earlier ones. The conversation you cleared is still saved: /resume opens it again. Files and checkpoints are not affected.

/stop

Stops the running task, the same as the Stop button or Esc. Whatever it finished stays, including edits already made; use /undo to reverse those. If nothing is running, the chat says so.

/settings

Opens AI settings, the same as the ⚙ button: provider and API key, command permissions, editor settings (autocomplete and language servers), MCP servers and codebase indexing.

/checkpoints and /undo

/checkpoints lists every restore point with its time; pick one to restore it. /undo restores the one saved before the latest task. Both show which files will change and ask you to confirm first. See Checkpoints and undo.

/resume

Conversations are saved on your computer as they happen, so closing Lamied, a crash or a power cut doesn't lose one. When you open a folder again, the conversation you had open there comes back. /resume lists the folder's other saved conversations, newest first, each named after its first request; pick one to continue it. Lamied keeps the 50 most recent conversations per folder.

/fork

Starts a new conversation from this one and leaves the original as it was, so you can try another approach without losing the first. /fork asks where to fork from:

• The whole conversation, the first choice: your next message continues a copy of it.
• Before one of your messages: the copy ends just before that message, and the message goes back into the message box so you can change it and send it again.

Give a name, as in /fork try-websockets, to tell the fork apart in /resume, where forks are marked Fork. Forking doesn't change any file: to put your files back as they were at that point, restore a checkpoint with /checkpoints. A fork starts with exactly the messages of the original, so a request sent within a few minutes can reuse the provider's prompt cache, which costs less.

/init

Writes AGENTS.md at the root of the workspace: how the project is laid out, the exact commands that build, test and run it, its code style, and its commit conventions. The assistant learns these from what the project declares (its manifests, CI files, README and git history), so the commands are real ones. The file is a project rule: every later request includes it, so the assistant doesn't have to work these things out again. Other coding agents read AGENTS.md too.

If the project already has AGENTS.md, LAMIED.md or .lamied/rules.md, the file is left alone, and the assistant says what it lacks instead.

/review

Reviews the changes since the last commit. Give a branch or commit to review against that instead, for example /review main. It's a review, not new work: the assistant reads and reports, and is told not to change any file.

• Findings come first, most severe first, each with its file and line: what is wrong, why it matters, and the fix.
• The review measures the time and space complexity of every changed function, as it was and as it is now (see Complexity analysis). A function that got slower, a new N+1 database query, or a new function at O(n²) or worse is reported with that evidence, not as an impression of the diff.
• Then come open questions, and a two-line summary of what the change does.

/help

Prints the list of commands, with their arguments, into the chat.