Code implemented tool calls

1 . A method, comprising:

receiving, at a server, a user request for execution of one or more tool calls;

generating, by a large language model (LLM), a code block in a programming language, the code block configured to encapsulate the one or more tool calls;

executing, by the server, the code block in a sandbox;

in response to obtaining a pending tool call, pausing execution of the code block;

transmitting the pending tool call to a client for execution;

receiving, from the client, a first result of the pending tool call;

resuming execution of the code block and substituting the first result of the pending tool call for the pending tool call in the code block; and

returning a second result of the executed code block to the LLM.

2 . The method of claim 1 , wherein at least one tool call of the one or more tool calls is designated for execution by the client.

3 . The method of claim 1 , wherein the sandbox is a resumable code execution sandbox.

4 . The method of claim 1 , wherein resuming the execution of the code block comprises using an evaluation stack that replays the code block from a beginning of the code block.

5 . The method of claim 4 , wherein the evaluation stack ensures deterministic execution by capturing and replaying results of non-deterministic operations.

6 . The method of claim 1 , wherein the LLM receives the second result and does not receive the first result.

7 . The method of claim 1 , wherein:

the code block supports parallel execution and sequential execution of the one or more tool calls;

the parallel execution is achieved using a parallel execution construct; and

the sequential execution is achieved using an await construct.

8 . The method of claim 1 , further comprising translating, by the server, tool definitions associated with the one or more tool calls from a JSON schema into a type definition in the programming language prior to generating the code block by the LLM.

9 . The method of claim 1 , wherein in response to determining an error in the execution of the one or more tool calls, capturing error information and providing the error information to the LLM.

10 . The method of claim 1 , further comprising wrapping non-deterministic operations in the code block with a function configured to store initial values of the non-deterministic operations.

11 . A computing system, comprising:

a processor; and

a memory storing instructions that, when executed by the processor, cause the computing system to:

receive, at a server, a user request for execution of one or more tool calls;

generate, by an LLM, a code block in a programming language, the code block configured to encapsulate the one or more tool calls;

execute, by the server, the code block in a sandbox;

in response to obtaining a pending tool call, pause execution of the code block;

transmit the pending tool call to a client for execution;

receive, from the client, a first result of the pending tool call;

resume execution of the code block and substituting the first result of the pending tool call for the pending tool call in the code block; and

return a second result of the executed code block to the LLM.

12 . The computing system of claim 11 , wherein at least one tool call of the one or more tool calls is designated for execution by the client.

13 . The computing system of claim 11 , wherein the sandbox is a resumable code execution sandbox.

14 . The computing system of claim 11 , wherein resume the execution of the code block comprises use an evaluation stack that replays the code block from a beginning of the code block.

15 . The computing system of claim 14 , wherein the evaluation stack ensures deterministic execution by capturing and replaying results of non-deterministic operations.

16 . The computing system of claim 11 , wherein the LLM receives the second result and does not receive the first result.

17 . The computing system of claim 11 , wherein:

the code block supports parallel execution and sequential execution of the one or more tool calls;

the parallel execution is achieved using a parallel execution construct; and

the sequential execution is achieved using an await construct.

18 . The computing system of claim 11 , wherein the computing system is further operable to translate, by the server, tool definitions associated with the one or more tool calls from a JSON schema into a type definition in the programming language prior to generating the code block by the LLM.

19 . The computing system of claim 11 , wherein in response to determining an error in the execution of the one or more tool calls, capture error information and providing the error information to the LLM.

20 . The computing system of claim 11 , wherein the computing system is further operable to wrap non-deterministic operations in the code block with a function configured to store initial values of the non-deterministic operations.