[Proposal] ToolChoice Ability and Expand Context Length

[haoruilee]

I am recently developing my OAOAgent Python client trying to finally build a general framework of on-chain AI agent, and while it has been extremely useful for interacting with AI oracles on the blockchain, I’m currently facing two limitations that I hope you could assist with:

• ToolChoice Ability for Models: At the moment, there are no LLMs with toolChoice ability deployed on OAO. This feature would significantly enhance the functionality and flexibility of AI agents by enabling them to perform more dynamic actions based on the user's input. Many agent framework like Swarm is based on this toolChoice ability. Could you consider deploying LLMs with toolChoice functionality to allow for more diverse and interactive agent behavior?

• Expanding Context Length: Currently, the context length is limited, which can be a significant constraint when managing longer conversations or more complex tasks. I’d appreciate it if there could be a way to expand the context to accommodate more extended interactions and retain relevant conversation history.

Looking forward to hearing your thoughts!

[0x1cc]

Nice work! Thanks for your interests in our OAO as well as the opAgent. We have released the latest version of opAgent. As for the limitations you have mentioned above.

• as for the ToolChoice, we have now supported the toolChoice functionality.
• For the offchain use case, the context length might be enough for most of the applications now. You can take a look on our RMS https://rms.ora.io/ For the onchain use case, the context length is limited due to the gas limit. It is infeasible to have a long context on the blockchain. But we will try to improve it.