Inception Labs Docs

Appearance

Next Edit ​

Fresh

Generate predictive code edit suggestions based on context: recently viewed files, current cursor position, and edit history.

Endpoint ​

POST https://api.inceptionlabs.ai/v1/edit/completions

Request Structure ​

The request content must include three sections:

1. Recently Viewed Snippets ​

Context from code the developer recently looked at. Include 3-5 snippets of roughly 20 lines each.

2. Current File Content ​

The file being edited, with the editable region marked using <|code_to_edit|> tags. Recommended: 10-15 lines (~100-150 tokens).

3. Edit Diff History ​

Recent changes in unidiff format. Include the last 3-5 edits, ordered chronologically (most recent last).

Example Request ​

bash
curl https://api.inceptionlabs.ai/v1/edit/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer $INCEPTION_API_KEY" \
  -d '{
    "model": "mercury-edit",
    "messages": [
      {
        "role": "user",
        "content": "Recently Viewed Snippets:\n...\n\nCurrent File: src/utils.js\n<|code_to_edit|>\nfunction processData(input) {\n  // TODO: implement\n}\n<|code_to_edit|>\n\nEdit History:\n..."
      }
    ],
    "max_tokens": 1000
  }'

Best Practices ​

ComponentRecommendation
Viewed Snippets3-5 snippets, ~20 lines each, centered on cursor positions
Edit Region10-15 lines (100-150 tokens), max ~25 lines
Region SelectionCenter around cursor: [currentLine - 5, currentLine + 10]
Edit HistoryLast 3-5 user edits in unidiff format, most recent last
Broader SuggestionsUse parallel requests with different editable regions

TIP

For broader edit suggestions, send parallel requests with different editable regions, or use linter-guided region selection.

See Also ​