Add fixes and todos

.env.example

@@ -1,10 +1,11 @@
 # Copy this file to .env and fill in the values
 
-## NOTE: The only required value is DEFAULT_OPENAI_API_KEY. The app should work if the other values 
+## NOTE: The only required values are DEFAULT_OPENROUTER_API_KEY and DEFAULT_OPENAI_API_KEY. The app should work if the other values 
 # are left as is or not defined at all. However, you are strongly encouraged to fill in your 
 # own SERPER_API_KEY value. It is also recommended to fill in the BYPASS_SETTINGS_RESTRICTIONS and
 # BYPASS_SETTINGS_RESTRICTIONS_PASSWORD values as needed (see below).
-DEFAULT_OPENAI_API_KEY="" # your OpenAI API key 
+DEFAULT_OPENROUTER_API_KEY="" # your OpenRouter API key 
+DEFAULT_OPENAI_API_KEY="" # your OpenAI API key
 
 ## Your Google Serper API key (for web searches). 
 # If you don't set this, my key will be used, which may have 
@@ -13,12 +14,12 @@ DEFAULT_OPENAI_API_KEY="" # your OpenAI API key
 # SERPER_API_KEY=""
 
 # Mode to use if none is specified in the query
-DEFAULT_MODE="/docs" # or "/web" | "/quotes" | "/details" | "/chat" | "/research"
+DEFAULT_MODE="/kb" # or "/chat", "/help", etc.
 
 # Variables controlling whether the Streamlit app imposes some functionality restrictions
 BYPASS_SETTINGS_RESTRICTIONS="" # whether to immediately allow all settings (any non-empty string means true)
-BYPASS_SETTINGS_RESTRICTIONS_PASSWORD="" # what to enter in the OpenAI API key field to bypass settings 
-# restrictions. If BYPASS_SETTINGS_RESTRICTIONS is non-empty or if the user enters their own OpenAI API key, 
+BYPASS_SETTINGS_RESTRICTIONS_PASSWORD="" # what to enter in the OpenRouter API key field to bypass settings 
+# restrictions. If BYPASS_SETTINGS_RESTRICTIONS is non-empty or if the user enters their own OpenRouter API key, 
 # this becomes - mostly - irrelevant, as full settings access is already granted. I say "mostly" because
 # this password can also be used for a couple of admin-only features, such as deleting the default collection
 # and deleting a range of collections (see dbmanager.py). Recommendation: for local use,
@@ -27,8 +28,8 @@ BYPASS_SETTINGS_RESTRICTIONS_PASSWORD="" # what to enter in the OpenAI API key f
 
 # If you are NOT using Azure, the following setting determines the chat model
 # NOTE: You can change the model and temperature on the fly in Streamlit UI or via the API
-MODEL_NAME="gpt-3.5-turbo-0125" # default model to use for chat
-ALLOWED_MODELS="gpt-3.5-turbo-0125, gpt-4-turbo-2024-04-09" 
+MODEL_NAME="google/gemini-2.5-flash" # default model to use for chat
+ALLOWED_MODELS="all" # Since we are now using OpenRouter, any model should be allowed. You can still specify a list if desired. 
 CONTEXT_LENGTH="16000" # you can also make it lower than the actual context length
 EMBEDDINGS_MODEL_NAME="text-embedding-3-large"
 EMBEDDINGS_DIMENSIONS="3072" # number of dimensions for the embeddings model

.gitignore

@@ -3,6 +3,8 @@ chroma-cloud*
 .tmp*
 credentials/
 chroma-cloud-backup/
+chroma/
+chroma.cf.json
 .chroma_env
 my-chroma*
 .chroma_env

.python-version

@@ -0,0 +1 @@
+3.13.3

README-FOR-DEVELOPERS.md

@@ -69,7 +69,7 @@ If this happens you will need to install the Microsoft C++ Build Tools. You can
 
 ### 4. Copy the `.env.example` file to `.env` and fill in the values
 
-At first, you can simply fill in your [OpenAI API key](https://platform.openai.com/signup) and leave the other values as they are. Please see `.env.example` for additional details.
+At first, you can simply fill in your [OpenRouter API key](https://openrouter.ai/CLERK-ROUTER/VIRTUAL/sign-up) and [OpenAI API key](https://platform.openai.com/signup) and leave the other values as they are. Please see `.env.example` for additional details.
 
 ## Running DocDocGo
 
@@ -116,6 +116,7 @@ The message should be sent as a POST request with the body as a JSON object that
 class ChatRequestData(BaseModel):
     message: str
     api_key: str
+    openrouter_api_key: str | None = None
     openai_api_key: str | None = None
     chat_history: list[JSONish] = []
     collection_name: str | None = None
@@ -146,7 +147,9 @@ The `collection_name` field is used to specify the collection that the bot shoul
 
 The `api_key` field is used to specify the API key for the FastAPI server. The server will only honor requests that include the correct API key, as specified by the `DOCDOCGO_API_KEY` environment variable in.
 
-The `openai_api_key` field is used to specify the OpenAI API key. If not specified, the default (community) key will be used, assuming the `DEFAULT_OPENAI_API_KEY` environment variable is set.
+The `openrouter_api_key` field is used to specify the OpenRouter API key. If not specified, the default (community) key will be used, assuming the `DEFAULT_OPENROUTER_API_KEY` environment variable is set.
+
+The `openai_api_key` field is used to specify the OpenAI API key for embeddings. If not specified, the default (community) key will be used, assuming the `DEFAULT_OPENAI_API_KEY` environment variable is set.
 
 The `access_codes_cache` field is an object mapping collection names to access codes that the client has stored for them for the current user. The bot will use these access codes to determine grant the user access to collections that require it.
 
@@ -392,7 +395,7 @@ As an alternative way to handle the issue of the default collection, you can cre
 
 ### Q: What is the `BYPASS_SETTINGS_RESTRICTIONS` environment variable?
 
-A: Normally, when this variable is not defined (or is an empty string), the app will start in a "community key" mode, where you can only see and create public collections and there are restriction on allowed settings (e.g. you can't change the model in the UI). The key used as the community key is controlled by the `DEFAULT_OPENAI_API_KEY` environment variable. You can remove these restrictions and switch to using that same key as a private key by entering the admin password (the value of the `BYPASS_SETTINGS_RESTRICTIONS_PASSWORD` environment variable) in rhe OpenAI API key field.
+A: Normally, when this variable is not defined (or is an empty string), the app will start in a "community key" mode, where you can only see and create public collections and there are restriction on allowed settings (e.g. you can't change the model in the UI). The keys used as the community keys are controlled by the `DEFAULT_OPENROUTER_API_KEY` and `DEFAULT_OPENAI_API_KEY` environment variables. You can remove these restrictions and switch to using those same keys as private keys by entering the admin password (the value of the `BYPASS_SETTINGS_RESTRICTIONS_PASSWORD` environment variable) in the OpenRouter API key field.
 
 However, when the `BYPASS_SETTINGS_RESTRICTIONS` variable is set to a non-empty string, the app will start in the "private key" mode right away, without you having to enter the admin password. This is useful if you use the app in a private setting and don't want to have to enter the admin password every time you start the app.
 

README.md

@@ -110,7 +110,7 @@ If this happens you will need to install the Microsoft C++ Build Tools. You can
 
 ### 4. Copy the `.env.example` file to `.env` and fill in the values
 
-At first, you can simply fill in your [OpenAI API key](https://platform.openai.com/signup) and leave the other values as they are. Please see `.env.example` for additional details.
+At first, you can simply fill in your [OpenRouter API key](https://openrouter.ai/CLERK-ROUTER/VIRTUAL/sign-up) and [OpenAI API key](https://platform.openai.com/signup) and leave the other values as they are. Please see `.env.example` for additional details.
 
 ## Running DocDocGo
 
@@ -431,9 +431,9 @@ A: Before you entered your own OpenAI API key, you were using the community key
 
 You still have access to the public collections, you can switch to any public collection by typing `/db use <collection name>`. If you want to see all available public collections again, you can switch back to the community key by changing the key to an empty string, then running `/db list` again.
 
-#### Q: I got a shareable link to a collection but using it reloads the Streamlit app, after which it ends up in its default state of using the community key. How can I use the link with my own OpenAI API key?
+#### Q: I got a shareable link to a collection but using it reloads the Streamlit app, after which it ends up in its default state of using the community key. How can I use the link with my own OpenRouter API key?
 
-A: Simply enter your key in the OpenAI API key field after the app has reloaded. The access code will still be valid.
+A: Simply enter your key in the OpenRouter API key field after the app has reloaded. The access code will still be valid.
 
 ## DocDocGo Carbon
 

agentblocks/webretrieve.py

@@ -17,6 +17,7 @@ class URLRetrievalData(BaseModel):
 
 
 MAX_INIT_BATCH_SIZE = 10
+MAX_URLS_TO_TRY = 25
 
 
 def get_content_from_urls(
@@ -31,8 +32,6 @@ def get_content_from_urls(
     Otherwise, fetch a new batch of urls, and repeat until at least min_ok_urls
     urls are fetched successfully.
 
-    If there are duplicate URLs
-
     Args:
     - urls: list of urls to fetch content from
     - min_ok_urls: minimum number of urls that need to be fetched successfully
@@ -42,11 +41,10 @@ def get_content_from_urls(
     Returns:
     - URLRetrievalData: object containing the fetched content
     """
+    urls = urls[:MAX_URLS_TO_TRY]
     try:
         batch_fetcher = batch_fetcher or get_batch_url_fetcher()
-        init_batch_size = init_batch_size or min(
-            MAX_INIT_BATCH_SIZE, round(min_ok_urls * 1.2)
-        )  # NOTE: could optimize
+        init_batch_size = init_batch_size or min(MAX_INIT_BATCH_SIZE, round(min_ok_urls * 1.2))  # NOTE: could optimize
 
         logger.info(
             f"Fetching content from {len(urls)} urls:\n"
@@ -85,11 +83,18 @@ def get_content_from_urls(
             batch_htmls = batch_fetcher(batch_urls)
 
             # Process fetched content
+            at_least_one_ok = False
             for url, html in zip(batch_urls, batch_htmls):
                 link_data = LinkData.from_raw_content(html)
                 res.link_data_dict[url] = link_data
                 if not link_data.error:
                     res.num_ok_urls += 1
+                    at_least_one_ok = True
+
+            if not at_least_one_ok:
+                errors = [res.link_data_dict[url].error for url in batch_urls]
+                error_string = "\n- ".join(f"{url}: {error}" for url, error in zip(batch_urls, errors))
+                logger.warning(f"No usable content found for any of the {batch_size} urls: {error_string}")
 
             logger.info(
                 f"Total URLs processed: {res.idx_first_not_tried} ({num_urls} total)\n"
@@ -98,6 +103,4 @@ def get_content_from_urls(
 
         return res
     except Exception as e:
-        raise DDGError(
-            user_facing_message="Apologies, I ran into a problem trying to fetch URL content."
-        ) from e
+        raise DDGError(user_facing_message="Apologies, I ran into a problem trying to fetch URL content.") from e