Uncensored AI for bioinformatics workflows
Glue code and parameter sanity for the pipeline you inherited
Half of bioinformatics is biology you can look up and half is shell scripting nobody wrote down. The second half is where the days go, and it is what a good model with tool calling is actually for.
Why the mainstream APIs fail here
The docs for the tool you are stuck on were last updated before the file format changed. A hosted model will discuss the biology happily and then hedge on the script, or the quota runs out mid-pipeline, and the run dies at 3am on the key you were given for the job.
How Wild West API is used for this
An OpenAI-compatible endpoint means the script calls the model the same way the tutorial said. Put a cap on the pipeline's own key so an overnight loop cannot spend past the budget. Ask for the parameter sanity check: depth, quality thresholds, what a failed QC looks like.
A working loop
- Give the model the command, the log and the file header, not the whole fastq.
- Ask what the parameter should be and why, then check the tool's own docs.
- Cap the key the cron job uses; the pipeline does not get to spend unbounded.
Details are on the privacy page. Per-model prices are on /models and the pricing model is on /pricing.