Our startup uses

langgraph

for building AI agentic workflows**. The team behind langgraph provide options for deploying agents in their cloud. Currently they only support the following two options: • Codebases thats use

requirements.txt

(docs). • Codebases that use

pyproject.toml

(docs). Our setup is a (1) monorepo, (2) we use protobuffs. I am looking into how I can make pants work under my constraints. ** Langgraph is the defacto framework for building AI agents.

So for most python workflows, Pants typically does stuff by creating a sandbox, copying and generating necessary files, and then performing some kind of action wrt to the sandbox. It sounds like you're ultimately trying to bypass sandboxing. Is that correct? Can you deploy from a sandbox via stuff like https://www.pantsbuild.org/2.21/docs/ad-hoc-tools/integrating-new-tools-without-plugins and run_shell_command?

Thanks. The

adhoc_tool

seem like something that could be useful. I could write a rule that copies the source files, run proto compiler, and generate the requirements.txt.

Ah sorry that was probably not a helpful. suggestion. The docs say

"description": "Whether to use a sandbox when `run`ning this target. Defaults to `[python].default_run_goal_use_sandbox`.\n\nIf true, runs of this target with the `run` goal will copy the needed first-party sources into a temporary sandbox and run from there.\n\nIf false, runs of this target with the `run` goal will use the in-repo sources directly.\n\nNote that this field only applies when running a target with the `run` goal. No other goals (such as `test`, if applicable) consult this field.\n\nThe former mode is more hermetic, and is closer to building and running the source as it were packaged in a `pex_binary`. Additionally, it may be necessary if your sources depend transitively on \"generated\" files which will be materialized in the sandbox in a source root, but are not in-repo.\n\nThe latter mode is similar to creating, activating, and using a virtual environment when running your files. It may also be necessary if the source being run writes files into the repo and computes their location relative to the executed files. Django's `makemigrations` command is an example of such a process.",

Looking at it now the "run" in the

default_run_goal_use_sandbox

really sticks out.