Server API

NovaEdit exposes a single edit endpoint over HTTP.

POST /v1/edit

Request

{
  "language": "python",
  "code": "...snippet...",
  "file_path": "app/routes.py",
  "start_line": 37,
  "end_line": 78,
  "diagnostics": ["NameError: name 'itm' is not defined at line 41"],
  "instruction": "fix errors only",
  "max_edits": 5,
  "temperature": 0.2
}

Response

{
  "edits": [
    {
      "start_line": 41,
      "end_line": 43,
      "replacement": "for i, item in enumerate(items):\n    process(item)\n"
    }
  ],
  "raw_patch_dsl": "@@ 41-43\n- for i in range(len(items)):\n-     process(items[i])\n+ for i, item in enumerate(items):\n+     process(item)\n",
  "model_version": "novaedit-baseline-0.1.0"
}

Running locally

uvicorn novaedit.server.main:app --reload --port 8000

Health check: GET /health → { "status": "ok", "version": "..." }

Configuration

• NOVAEDIT_MODEL_ID — optional HF model ID to load (default is heuristic baseline).
• NOVAEDIT_DEVICE — device string (e.g., cuda:0).
• NOVAEDIT_LANGUAGE — default python (javascript stub also wired).
• NOVAEDIT_MAX_CODE_LINES — reject snippets above this line count (default 2000).
• NOVAEDIT_MAX_CONCURRENT — reject requests over this concurrency (default 8).
• NOVAEDIT_REQUEST_TIMEOUT — seconds before timing out a request (default 15).
• NOVAEDIT_LOG_REQUESTS — set to true to log edit calls.
• NOVAEDIT_CORS_ORIGINS — comma-separated list of allowed origins (add if calling from browser plugins).

Error handling

• 400 if start_line > end_line or payload is invalid.
• 200 with zero edits if the model emits no changes.

Notes

• Current model is heuristic; replace NovaEditModel with a trained checkpoint to upgrade quality.
• Patch DSL is line-based; the server converts it to structured edits in JSON for clients.