diff --git a/app/api/routes/forms.py b/app/api/routes/forms.py
index be16029..caab2a7 100644
--- a/app/api/routes/forms.py
+++ b/app/api/routes/forms.py
@@ -1,7 +1,9 @@
 from datetime import datetime, timedelta, timezone
 from pathlib import Path
+import json
 import requests
 from fastapi import APIRouter, Depends, File, UploadFile, Query
+from fastapi.responses import StreamingResponse
 from sqlmodel import Session, select
 
 from app.api.deps import get_db, verify_api_key
@@ -10,6 +12,7 @@
     FormFillResponse,
     ModelsResponse,
     TranscriptionResponse,
+    ModelPullRequest,
 )
 from app.core.config import OLLAMA_HOST, OLLAMA_MODEL, BASE_DIR, RETENTION_PERIOD_DAYS
 from app.services.whisper import call_whisper_asr
@@ -86,6 +89,33 @@ def list_models():
     return ModelsResponse(models=models, default=default_model)
 
 
+@router.post("/pull")
+def pull_model(req: ModelPullRequest):
+    """Stream pull progress from Ollama as NDJSON.
+    Each line is a JSON object with Ollama's progress fields
+    (status, completed, total, digest). The last line will
+    have status='success' when the pull completes."""
+
+    def generate():
+        try:
+            with requests.post(
+                f"{OLLAMA_HOST}/api/pull",
+                json={"name": req.model, "stream": True},
+                stream=True,
+                timeout=600,
+            ) as r:
+                r.raise_for_status()
+                for chunk in r.iter_content(chunk_size=None):
+                    if chunk:
+                        yield chunk
+            # Signal clean completion
+            yield (json.dumps({"status": "success"}) + "\n").encode("utf-8")
+        except requests.exceptions.RequestException as e:
+            yield (json.dumps({"error": str(e)}) + "\n").encode("utf-8")
+
+    return StreamingResponse(generate(), media_type="application/x-ndjson")
+
+
 @router.post("/transcribe", response_model=TranscriptionResponse)
 def transcribe(audio: UploadFile = File(...)):
     """Forward recorded audio to the local Whisper ASR sidecar and return text.
diff --git a/app/api/schemas/forms.py b/app/api/schemas/forms.py
index 155b14f..5acdadc 100644
--- a/app/api/schemas/forms.py
+++ b/app/api/schemas/forms.py
@@ -74,4 +74,8 @@ class Config:
 
 
 class AsyncFormFillResponse(BaseModel):
-    jobs: list[AsyncJobSubmitResponse]
\ No newline at end of file
+    jobs: list[AsyncJobSubmitResponse]
+
+
+class ModelPullRequest(BaseModel):
+    model: str
\ No newline at end of file
diff --git a/docs/1. SETUP.md b/docs/1. SETUP.md
index 30d2aee..23d3ecc 100644
--- a/docs/1. SETUP.md	
+++ b/docs/1. SETUP.md	
@@ -91,6 +91,19 @@ Check `make logs-app` for the actual error. The entrypoint runs database migrati
 **Want a clean slate**
 `make super-clean` stops everything and **deletes all volumes** database, uploads, and downloaded model weights. Only use it when you intend to wipe all local data.
 
+## AI Model Selection
+
+FireForm allows you to choose which AI model to run during form extraction directly from the dropdown in the frontend "Fill Form" UI.
+
+The supported recommended models are:
+- `qwen2.5:1.5b` (default, lightweight)
+- `qwen2.5:3b`
+- `qwen2.5:7b`
+- `llama3.2:3b`
+- `mistral:7b`
+
+If you select a model that is not yet installed (pulled) in your local Ollama instance, the app will automatically request the backend to download it via the `POST /api/v1/forms/pull` endpoint. During installation, form submission is temporarily disabled, and status progress is displayed. Once downloaded, the model is cached in Ollama's Docker volume for future use.
+
 ## Where to go next
 
 - **Join our [Discord](https://discord.gg/nBv5b6kF68)** — ask questions and coordinate with other contributors
diff --git a/tests/test_api.py b/tests/test_api.py
index 32104ae..0b28e79 100644
--- a/tests/test_api.py
+++ b/tests/test_api.py
@@ -283,6 +283,34 @@ def boom(*a, **k):
         assert resp.status_code == 200
         assert resp.json()["models"] == ["qwen2.5:1.5b"]
 
+    def test_pull_model_success(self, client, monkeypatch):
+        from unittest.mock import MagicMock
+        fake_response = MagicMock()
+        fake_response.raise_for_status.return_value = None
+        monkeypatch.setattr("app.api.routes.forms.requests.post", lambda *a, **k: fake_response)
+
+        resp = client.post(f"{API_PREFIX}/forms/pull", json={"model": "llama3.2:3b"})
+        assert resp.status_code == 200
+        assert resp.json()["status"] == "success"
+
+    def test_pull_model_failure(self, client, monkeypatch):
+        """When Ollama is unreachable the stream returns 200 with an error line in the body."""
+        import requests
+        import json
+
+        def boom(*a, **k):
+            raise requests.exceptions.RequestException("pull failed")
+
+        monkeypatch.setattr("app.api.routes.forms.requests.post", boom)
+
+        resp = client.post(f"{API_PREFIX}/forms/pull", json={"model": "llama3.2:3b"})
+        assert resp.status_code == 200
+        # The streamed body contains an error JSON line
+        lines = [l for l in resp.text.splitlines() if l.strip()]
+        assert lines, "expected at least one NDJSON line"
+        last = json.loads(lines[-1])
+        assert "error" in last
+
     def test_fill_form_passes_model_override(self, client, mock_controller):
         """A `model` in the request reaches Controller.fill_form but isn't persisted."""
         tpl_id = self._seed_template(client, mock_controller)