Autocomplete
Inline AI suggestions while you type, and choosing their model with /fim.
As you type, Lamied suggests how to continue your code in grey “ghost text”. It's on by default for Pro accounts.
Accept or dismiss
| To… | Press |
|---|---|
| Accept the whole suggestion | Tab |
| Accept just the next word | Ctrl→ |
| Dismiss it | Esc |
| Ask for a suggestion now, from your chat model | Alt\ |
Or keep typing: the suggestion updates or disappears.
When suggestions appear
- About a third of a second after you pause typing.
- At the end of a line, or where only closing characters like
),],}or;follow the cursor. - In the middle of a line, only the rest of that line is suggested.
- At the end of a line, the statement or block you're starting. After
class Account:orif (a) {, it suggests the whole body, and stops where the block ends. - On a blank line, the suggestion brings its own correct indentation.
Turn it on or off
Open AI settings (⚙ or /settings) → Editor → Inline AI suggestions while typing. The choice is remembered on this computer.
Choose the model
Autocomplete uses its own model setting, separate from the chat. Type /fim to choose it, or /fim auto to return to automatic.
| Setting | Model used while typing |
|---|---|
| Automatic, with a Mistral key saved | Codestral, a model built for completing code in the middle of a file |
| Automatic, Gemini as chat provider | Gemini's fast Flash-Lite model |
| Automatic, Lamied (Team plan) as chat provider | Gemini's fast Flash-Lite model through Lamied, counted against your team's allowance |
| Automatic, other providers | Your selected chat model |
/fim choice | Exactly the provider and model you picked |
Speed matters most here: small models built for speed answer quickly, while large models can take seconds per suggestion. If suggestions feel slow, choose a small, fast model with /fim. Every pause in typing is one request, so with your own key a large model here costs far more than in the chat.
What is sent, and what never is
At each pause, the code around your cursor (up to about 6,000 characters before it and 2,000 after), the file's name without its folder path, and its language go to the autocomplete model. Nothing is sent while autocomplete is off or your account isn't Pro.
Secret files are never sent. Typing in .env files, private keys and certificates (*.pem, *.key, id_rsa…), .npmrc, .pypirc, .netrc, .git-credentials or credentials files gets no suggestions and makes no request. Templates like .env.example are allowed. The same rule keeps these files' contents away from the AI assistant.
If suggestions stop
If the provider rejects a request (a wrong key, no access to the model, or a rate limit), suggestions pause for 15 seconds instead of repeating the error. Alt\ still works during the pause. Check your key in settings, or pick another model with /fim. See Troubleshooting.