gavin/maybe
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases 28 Wiki Activity

Get assistant test working again

Browse Source
This commit is contained in:
Zach Gollwitzer
2025-03-21 17:20:18 -04:00
parent 63e24623fd
commit 8e698bfd07
5 changed files with 56 additions and 69 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
app/models/assistant.rb
View File

@@ -10,12 +10,12 @@ class Assistant
def available_functions
[
Assistant::Function::GetBalanceSheet,
Assistant::Function::GetIncomeStatement,
Assistant::Function::GetExpenseCategories,
Assistant::Function::GetAccountBalances,
Assistant::Function::GetTransactions,
Assistant::Function::ComparePeriods
Assistant::Functions::GetBalanceSheet,
Assistant::Functions::GetIncomeStatement,
Assistant::Functions::GetExpenseCategories,
Assistant::Functions::GetAccountBalances,
Assistant::Functions::GetTransactions,
Assistant::Functions::ComparePeriods
]
end
@@ -72,19 +72,18 @@ class Assistant
return
end
message = response.data.message
message.chat = chat
message.status = "pending"
# If no tool calls, create a plain message for the chat
unless response.data.tool_calls.any?
message = response.data.message
message.status = "complete"
message.save!
return
end
# Step 1: Saving a "pending" message with incomplete tool call definitions
message = response.data.message
message.status = "pending"
message.save!
# Step 2: Call the functions, add to message and save
# Step 1: Call the functions, add to message and save
tool_calls = message.tool_calls.map do |tool_call|
result = call_tool_function(tool_call.function_name, tool_call.function_arguments)
tool_call.function_result = result
@@ -94,7 +93,7 @@ class Assistant
message.tool_calls = tool_calls
message.save!
# Step 3: Call LLM again with tool call results and update the message with response
# Step 2: Call LLM again with tool call results and update the message with response
second_response = provider.chat_response(
model: latest_message.ai_model,
instructions: instructions,
@@ -107,9 +106,10 @@ class Assistant
return
end
second_message = second_response.data.message
second_message.status = "complete"
second_message.save!
# Step 3: Update the message with the final response
message.status = "complete"
message.content = second_response.data.message.content
message.save!
end
private
@@ -117,10 +117,10 @@ class Assistant
chat.messages.ordered.where(role: [ :user, :assistant, :developer ], status: "complete", kind: "text")
end
def call_tool_function(fn_name, fn_params = {})
def call_tool_function(fn_name, fn_params)
fn = available_functions.find { |fn| fn.name == fn_name }
raise "Assistant does not implement function: #{fn_name}" if fn.nil?
fn.call(fn_params)
fn.call(JSON.parse(fn_params))
end
def instructions

13
app/models/provider/openai.rb
View File

@@ -24,7 +24,6 @@ class Provider::OpenAI < Provider
message: Message.new(
ai_model: response.dig("model"),
provider_id: response.dig("id"),
status: normalize_status(response.dig("status")),
role: "assistant",
content: extract_content(response),
),
@@ -89,16 +88,4 @@ class Provider::OpenAI < Provider
)
end
end
# Normalize to our internal message status values
def normalize_status(status)
case status
when "in_progress"
"pending"
when "completed"
"complete"
when "incomplete"
"failed"
end
end
end

14
test/models/assistant_test.rb
View File

@@ -14,13 +14,13 @@ class AssistantTest < ActiveSupport::TestCase
@provider.expects(:chat_response).returns(
provider_success_response(
Assistant::Provideable::ChatResponse.new(
messages: [
Message.new(
role: "assistant",
content: "Hello from assistant",
ai_model: "gpt-4o"
)
]
message: Message.new(
chat: @chat,
role: "assistant",
content: "Hello from assistant",
ai_model: "gpt-4o"
),
tool_calls: []
)
)
)

6
test/models/provider/openai_test.rb
View File

@@ -25,7 +25,7 @@ class Provider::OpenAITest < ActiveSupport::TestCase
test "handles chat response with tool calls" do
VCR.use_cassette("open_ai/chat/tool_calls", record: :all) do
class TestFn
class PredictableToolFunction
include Assistant::Functions::Toolable
class << self
@@ -48,7 +48,7 @@ class Provider::OpenAITest < ActiveSupport::TestCase
response = @openai.chat_response(
model: "gpt-4o",
instructions: Assistant.instructions,
functions: [ TestFn ],
functions: [ PredictableToolFunction ],
messages: [ initial_message ]
)
@@ -56,7 +56,7 @@ class Provider::OpenAITest < ActiveSupport::TestCase
assert response.data.tool_calls.size == 1
tool_call = response.data.tool_calls.first
tool_call_result = TestFn.new.call(JSON.parse(tool_call.function_arguments))
tool_call_result = PredictableToolFunction.new.call(JSON.parse(tool_call.function_arguments))
message_with_tool_calls = Message.new(
role: "assistant",

54
test/vcr_cassettes/open_ai/chat/tool_calls.yml
View File

@@ -35,7 +35,7 @@ http_interactions:
message: OK
headers:
Date:
- Fri, 21 Mar 2025 20:56:14 GMT
- Fri, 21 Mar 2025 21:09:55 GMT
Content-Type:
- application/json
Transfer-Encoding:
@@ -47,34 +47,34 @@ http_interactions:
Openai-Organization:
- "<OPENAI_ORGANIZATION_ID>"
X-Request-Id:
- req_d0dcb7a2bb6b188cc992f81b1171ec71
- req_df82a0b6c7e52c617bea4a7f255cf414
Openai-Processing-Ms:
- '807'
- '1243'
Strict-Transport-Security:
- max-age=31536000; includeSubDomains; preload
Cf-Cache-Status:
- DYNAMIC
Set-Cookie:
- __cf_bm=mm1LOM_CqqDbpNxg5U9POkF8mZmbwy93TakM0UNW79Y-1742590574-1.0.1.1-xEzx9bxl_Ql_u0SX6Artx49KLfEaj2odnlOpyzz8igb8wqVDvALU53jeepQtphRu53x4gCnq6Vafxmchv7oh3nb36_iH_i5kU105C10gfyk;
path=/; expires=Fri, 21-Mar-25 21:26:14 GMT; domain=.api.openai.com; HttpOnly;
- __cf_bm=DarUkvh8DnHC5TCIieafszl9rDHMrLr.4cVMzAWF5l8-1742591395-1.0.1.1-YQVP1gM3xI31iyBH7ou1ojCE8yUP31p1fKaJQBZM6fAXZM_Z_N5SYYXUJkiEqYCeZ84nXB.UpZRIQ7lwngV7fYLOyaAky4mru5GIwWrsRSs;
path=/; expires=Fri, 21-Mar-25 21:39:55 GMT; domain=.api.openai.com; HttpOnly;
Secure; SameSite=None
- _cfuvid=OmEvdCnIjJ7f7pONp9es_4f0YJTR7ZzOa5JffY6t7.8-1742590574225-0.0.1.1-604800000;
- _cfuvid=Ka1bmyiJo_efDQsHH76aUZgiza8FUt0vjocA0IRkM6Q-1742591395414-0.0.1.1-604800000;
path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None
X-Content-Type-Options:
- nosniff
Server:
- cloudflare
Cf-Ray:
- 92405acb3c99cf56-CMH
- 92406ed51b76f60e-ORD
Alt-Svc:
- h3=":443"; ma=86400
body:
encoding: ASCII-8BIT
string: |-
{
"id": "resp_67ddd26d66e081928243bb73ccf4f17d051b843636768b97",
"id": "resp_67ddd5a229048192959a8300ba51b0bd07102ebe9bb01597",
"object": "response",
"created_at": 1742590573,
"created_at": 1742591394,
"status": "completed",
"error": null,
"incomplete_details": null,
@@ -84,8 +84,8 @@ http_interactions:
"output": [
{
"type": "function_call",
"id": "fc_67ddd26de5f881929ee7297fb8cee1db051b843636768b97",
"call_id": "call_58gKqckPCeWSPwYdbP3EBpAY",
"id": "fc_67ddd5a33d9c8192a9df6aa1232dc58607102ebe9bb01597",
"call_id": "call_usxRDYgph9du3IYoY2T4VDvM",
"name": "get_net_worth",
"arguments": "{}",
"status": "completed"
@@ -135,14 +135,14 @@ http_interactions:
"user": null,
"metadata": {}
}
recorded_at: Fri, 21 Mar 2025 20:56:14 GMT
recorded_at: Fri, 21 Mar 2025 21:09:55 GMT
- request:
method: post
uri: https://api.openai.com/v1/responses
body:
encoding: UTF-8
string: '{"model":"gpt-4o","input":[{"role":"user","content":"What is my net
worth?"},{"type":"function_call","id":"fc_67ddd26de5f881929ee7297fb8cee1db051b843636768b97","call_id":"call_58gKqckPCeWSPwYdbP3EBpAY","name":"get_net_worth","arguments":"{}"},{"type":"function_call_output","call_id":"call_58gKqckPCeWSPwYdbP3EBpAY","output":"$124,200"}],"tools":[],"instructions":"You
worth?"},{"type":"function_call","id":"fc_67ddd5a33d9c8192a9df6aa1232dc58607102ebe9bb01597","call_id":"call_usxRDYgph9du3IYoY2T4VDvM","name":"get_net_worth","arguments":"{}"},{"type":"function_call_output","call_id":"call_usxRDYgph9du3IYoY2T4VDvM","output":"$124,200"}],"tools":[],"instructions":"You
are a helpful financial assistant for Maybe, a personal finance app.\nYou
help users understand their financial data by answering questions about their
accounts, transactions, income, expenses, and net worth.\n\nWhen users ask
@@ -170,7 +170,7 @@ http_interactions:
message: OK
headers:
Date:
- Fri, 21 Mar 2025 20:56:15 GMT
- Fri, 21 Mar 2025 21:09:56 GMT
Content-Type:
- application/json
Transfer-Encoding:
@@ -182,34 +182,34 @@ http_interactions:
Openai-Organization:
- "<OPENAI_ORGANIZATION_ID>"
X-Request-Id:
- req_f3be656bce149275f2368834bfd83be9
- req_5b8e1f5ed4334f5e8db29077290b26ec
Openai-Processing-Ms:
- '632'
- '693'
Strict-Transport-Security:
- max-age=31536000; includeSubDomains; preload
Cf-Cache-Status:
- DYNAMIC
Set-Cookie:
- __cf_bm=.lL8VlbRrdmeXay2X_ltzgMNGI5tbIvf.LUlQku9BRE-1742590575-1.0.1.1-2c6E2_eVRbrOSWoqnj3ROv59Ay1i9fc.dNT5rDGXOjyjISNTsJSeExl3lnng.4MgQUx0nzt5xAeH3kHonwDS4nNQtHJGtIx2Vyc1qiV.zHk;
path=/; expires=Fri, 21-Mar-25 21:26:15 GMT; domain=.api.openai.com; HttpOnly;
- __cf_bm=Kf9FcHhzonwgoRJaHjiOGgm7cD3Z477g42Tesl2ZJb0-1742591396-1.0.1.1-CzEx76PfoASk87nLiy7xodLyELqvlu6ajZCodw9WvO95K776mh7.AsM9CW7uRXt_rtc5D9QrwzdTAJKZA7jQDndT_CcWhT1V_bBTwPdr6mA;
path=/; expires=Fri, 21-Mar-25 21:39:56 GMT; domain=.api.openai.com; HttpOnly;
Secure; SameSite=None
- _cfuvid=Puw5FtUzW08qv5IvAztCHaMlK0JB3huEVX8tx_G7wJ4-1742590575078-0.0.1.1-604800000;
- _cfuvid=aNVZsrX3xANqvEFRKpN0pc8S5H2V1LYEYIX0hU_eRCY-1742591396329-0.0.1.1-604800000;
path=/; domain=.api.openai.com; HttpOnly; Secure; SameSite=None
X-Content-Type-Options:
- nosniff
Server:
- cloudflare
Cf-Ray:
- 92405ad1b9fecf5e-CMH
- 92406ede0db0cf46-CMH
Alt-Svc:
- h3=":443"; ma=86400
body:
encoding: ASCII-8BIT
string: |-
{
"id": "resp_67ddd26e6c048192bb1e41606aa6b68e051b843636768b97",
"id": "resp_67ddd5a39bf48192a536426c7c237b6307102ebe9bb01597",
"object": "response",
"created_at": 1742590574,
"created_at": 1742591395,
"status": "completed",
"error": null,
"incomplete_details": null,
@@ -219,13 +219,13 @@ http_interactions:
"output": [
{
"type": "message",
"id": "msg_67ddd26eb1748192b9f506f5cdd73300051b843636768b97",
"id": "msg_67ddd5a3ef28819290093f3b08ecdc3207102ebe9bb01597",
"status": "completed",
"role": "assistant",
"content": [
{
"type": "output_text",
"text": "Your net worth is **$124,200**.\n\nWould you like to see a breakdown of your assets and liabilities?",
"text": "Your net worth is **$124,200**.\n\nWant to explore your assets or liabilities?",
"annotations": []
}
]
@@ -253,14 +253,14 @@ http_interactions:
"input_tokens_details": {
"cached_tokens": 0
},
"output_tokens": 25,
"output_tokens": 20,
"output_tokens_details": {
"reasoning_tokens": 0
},
"total_tokens": 228
"total_tokens": 223
},
"user": null,
"metadata": {}
}
recorded_at: Fri, 21 Mar 2025 20:56:15 GMT
recorded_at: Fri, 21 Mar 2025 21:09:56 GMT
recorded_with: VCR 6.3.1