I tried it now, wonderful interface, but I see that it has the same issue i m facing, which is once you give it a large text, it is not able to process it, and start bubbling about how you did not give it data to do its job based on the instructions, what do you think this issue is related to? capability of assistant, because I tried all models, all the same within assistants.
Your Chatbot UI is amazing, thanks for sharing.
Question please, I m able to make it work in my solution, but it’s all over the place, one time it returns the message, another it returns the second message, not the first, so I might be doing wrong, here is my outline and I m no coder, i m using a no-code solution to create a conversation Assistant in employee performance management, so here is my layout, what I am I doing wrong?
- Post / I m creating thread using this : https://api.openai.com/v1/threads/runs and obviously using the assistant ID, which works I can see the thread in OpenAI
- Post/ I m pulling the messages through this : https://api.openai.com/v1/threads/{threadID}/messages/
- Post/ https://api.openai.com/v1/threads/{threadID}/runs with JSON reference to the assistant ID
- GET/ https://api.openai.com/v1/threads/{threadID}/messages/ to get the response…
What am I doing wrong, sorry for the format, I m no coder but I survive lol, using Adalo for the custom action.
Thank you
I’m assuming since there is no way to manage the messages contained in specific threads, that we will need to keep track of messages and threads on our own, potentially using a manual rating system for message importance or using concepts like pinned messages, and then dynamically create new threads with pertinent messages for specific runs.
Has anyone implemented or tried anything like this yet?
If you try to write new threads, you cannot place the messages and responses from previous tool calls. That will mean the AI cannot continue upon or make decisions on prior interactions with code interpreter, knowledge, your tools, or even understand why it answered the way it did.
You have an API parameter now to limit the number of past chat turns used in a run, but they don’t specify exactly how it would apply to each message by type: truncation_strategy → last_messages or if it affects continuing internal model calls.
I can’t manually copy the contents of a previous thread and insert it into a new thread?
Only the user and plain text assistant messages can be retrieved and placed.
If the last turn was:
user: do salts give us a path to fusion instead of fission?
assistant: to= file_search({“query”:“salt reactor nuclear fission”})
tool: (16kB of vector retrieval about search terms.
assistant: I’ve searched, and, nope, buddy.
Then you lose the middle steps and the documentation previously employed.
You can’t ask “what was the exact error in posting the tweet?”
1 Like
I’m honestly shocked to hear this. This really handicaps the entire API. Reading through the docs, I thought there must be some workflow which would allow for a bit more control and customization, especially considering the pricing and context situation, but apparently that is not the case… what types of strategies have people been able to come up with so far?
I can still the utility in Agents as they are currently, but until developers are given more control (just a little would be nice) over Thread/Run management, this API is bricked imo.
1 Like
thanks
I don’t unserstand why would we create an unique assitant for every user?
Can’t same assistant be user for all the users?
You can limit the context tokens