AI tools at BioHPC
This page provides our experience and recommendations for using available
AI tools effectively in the BioHPC environment.
The first step in using AI tools efficiently is to give them enough context
about how BioHPC works. This includes what software is available, where and
how programs should be run, and where users can find additional documentation.
For example, Docker is available
in BioHPC through the docker1 command.
If your AI does not know this, it may try the standard docker
command, see that it fails, and incorrectly assume that Docker is not available.
This can lead to unnecessary attempts to install software or configure tools that are already present.
Another common issue involves temporary files. Some pipelines create large
temporary files and may try to write them to /tmp.
On BioHPC systems, this can overload the root filesystem.
Temporary files should instead be written to: /workdir/$USER.
Using AI in Chat mode
If you use AI in chat mode, you can download this
guardrail file to teach it about BioHPC.
The file is in Markdown format (.md) format.
The
easiest way to use it in chat mode is to upload the file to the chat and
tell AI something like:
Review the attached guardrail file. For the rest of this conversation, follow these rules.
Some AI tools also allow you to include these instructions automatically in future chats,
although the method depends on the tool. For example, ChatGPT recommends
Custom GPT for this purpose.
Using AI agents in a linux environment
If you are using an AI agent in a Linux environment, you can copy the appropriate guardrail file to the agent's configuration directory.
For Codex:
mkdir -p /home/$USER/.codex
cp
/programs/ai_pipelines/AGENTS.md /home/$USER/.codex
For Claude:
mkdir -p /home/$USER/.claude
cp
/programs/ai_pipelines/CLAUDE.md /home/$USER/.claude
In April 2026 we presented an AI-themed workshop "
From
ChatGPT to AI Agents: Practical RNA-seq Workflows with Codex CLI
",
which you may find useful.