diff --git a/src/foiaghost/ghost.py b/src/foiaghost/ghost.py
new file mode 100644
index 0000000..e167158
--- /dev/null
+++ b/src/foiaghost/ghost.py
@@ -0,0 +1,28 @@
+from scrapeghost import SchemaScraper
+
+schema = {
+    "public_records_email": "email",
+    "public_records_address": "str",
+    "public_records_phone": "555-555-5555",
+    "public_records_fax": "555-555-5555",
+    "public_records_web": "url",
+    "general_contact_phone": "555-555-5555",
+    "general_contact_address": "str",
+    "foia_guide": "url",
+    "public_reading_room": "url",
+    "agency_logo": "url",
+}
+extra_instructions = """
+The fields that begin with public_records should refer to contact information specific to FOIA/Public Information/Freedom of Information requests.
+The fields that begin with general_contact should refer to contact information for the agency in general.
+If a field is not found in the HTML, leave it as null in the JSON.
+"""
+
+# create a scrapeghost
+ghost = SchemaScraper(
+    schema=schema,
+    models=["gpt-3.5-turbo-16k"],
+    # extra_preprocessors=[],
+    extra_instructions=extra_instructions,
+    max_cost=10,
+)
diff --git a/src/foiaghost/models.py b/src/foiaghost/models.py
index 61363e6..2d8cdce 100644
--- a/src/foiaghost/models.py
+++ b/src/foiaghost/models.py
@@ -13,3 +13,17 @@ class URL(BaseModel):
 
 class Int(BaseModel):
     int: int
+
+
+class IdOnly(BaseModel):
+    pass
+
+
+class JSON(BaseModel):
+    scraped_json: dict | list
+
+
+class ScrapeghostResponse(BaseModel):
+    total_cost: float
+    api_time: float
+    data: dict
diff --git a/src/foiaghost/pipeline.py b/src/foiaghost/pipeline.py
index 22fbf35..9864574 100644
--- a/src/foiaghost/pipeline.py
+++ b/src/foiaghost/pipeline.py
@@ -1,14 +1,19 @@
-from ssl import SSLCertVerificationError, SSLError
+import csv
 import httpx
 import tiktoken
 import lxml.html
+import lxml.etree
 from lxml.etree import ParserError
-from databeakers import Pipeline
+from ssl import SSLCertVerificationError, SSLError
+from databeakers.pipeline import Pipeline, EdgeType, ErrorType
+from databeakers.beakers import TempBeaker
 from databeakers.http import HttpRequest, HttpResponse
-from scrapeghost import SchemaScraper
 from scrapeghost.preprocessors import CleanHTML
-from .models import Agency, URL, Int
-import csv
+from scrapeghost.errors import TooManyTokens
+from openai.error import InvalidRequestError
+from scrapeghost.errors import BadStop
+from .models import Agency, URL, Int, IdOnly, ScrapeghostResponse
+from .ghost import ghost
 
 
 class CSVSource:
@@ -24,10 +29,10 @@ class CSVSource:
 
 
 def tiktoken_count(response):
-    if response["status_code"] != 200:
+    if response.status_code != 200:
         raise ValueError("response status code is not 200")
 
-    html = response["response_body"]
+    html = response.response_body
 
     # clean the html
     cleaner = CleanHTML()
@@ -37,50 +42,69 @@ def tiktoken_count(response):
     html_again = lxml.html.tostring(doc, encoding="unicode")
     tokens = len(encoding.encode(html_again))
 
-    response["tiktoken_count"] = tokens
-
-    return response
+    return Int(int=tokens)
 
 
 # current thinking, beakers exist within a recipe
 recipe = Pipeline("foiaghost", "foiaghost.db")
 recipe.add_beaker("agency", Agency)
-recipe.add_beaker("good_urls", URL)
-recipe.add_transform("agency", "good_urls", lambda x: x["url"].startswith("http"))
-recipe.add_beaker("responses", HttpResponse)
-recipe.add_transform("good_urls", "responses", HttpRequest)
-recipe.add_beaker("tiktoken_count", Int)
-recipe.add_transform(
-    "responses",
-    "tiktoken_count",
-    tiktoken_count,
-    error_map={(ValueError, ParserError): "no_tiktoken_count"},
-)
 recipe.add_seed(
     "agencies",
     "agency",
     CSVSource("agencies.csv", Agency),
 )
-
-# recipe.add_beaker("token_lt_8k", temp=True)
-# recipe.add_beaker("token_gt_8k", temp=True)
+recipe.add_beaker("good_urls", URL)
+recipe.add_transform(
+    "agency",
+    "good_urls",
+    lambda a: a.url.startswith("http"),
+    edge_type=EdgeType.conditional,
+)
+recipe.add_beaker("responses", HttpResponse)
+recipe.add_transform(
+    "good_urls",
+    "responses",
+    HttpRequest(),
+    error_map={
+        (
+            httpx.HTTPError,
+            SSLCertVerificationError,
+            SSLError,
+        ): "bad_requests"
+    },
+)
 
 
-# recipe.add_transform(
-#     "good_urls",
-#     "responses",
-#     add_response,
-#     error_map={
-#         (
-#             httpx.HTTPError,
-#             SSLCertVerificationError,
-#             SSLError,
-#         ): "bad_requests"
-#     },
-# )
-# recipe.add_conditional(
-#     "with_tiktoken_count",
-#     lambda x: x["tiktoken_count"] < 8000,
-#     if_true="token_lt_8k",
-#     if_false="token_gt_8k",
-# )
+class ProcessRecord:
+    def __init__(self, func, params_map):
+        self.func = func
+        self.params_map = params_map
+
+    def __call__(self, record):
+        kwargs = {}
+        for param, (beaker_name, field_name) in self.params_map.items():
+            kwargs[param] = getattr(record[beaker_name], field_name)
+        return self.func(**kwargs)
+
+    def __repr__(self):
+        return f"ProcessRecord({self.func.__name__}, {self.params_map})"
+
+
+def scrapeghost_response(record) -> ScrapeghostResponse:
+    sg = ghost.scrape(url_or_html=record["responses"].response_body)
+    return ScrapeghostResponse(
+        total_cost=sg.total_cost, api_time=sg.api_time, data=sg.data
+    )
+
+
+recipe.add_transform(
+    "responses",
+    "scrapeghost_response",
+    scrapeghost_response,
+    whole_record=True,
+    error_map={
+        (TooManyTokens,): "scrapeghost_too_many_tokens",
+        (BadStop,): "scrapeghost_bad_stop",
+        (InvalidRequestError, ValueError, ParserError): "scrapeghost_invalid_request",
+    },
+)