Product · In development
Deterministic pipeline engineering for Nextflow.
Codaris is a development environment for Nextflow DSL2 and nf-core pipelines. You describe the analysis; deterministic code scaffolds the project, resolves modules and containers, wires channels, lints against nf-core rules and runs a local test profile. The language model handles only small, validator-gated tasks, so hallucinated container tags and broken channel wiring are caught before anything reaches your cluster.
01 / What it does
Capabilities.
- 01
nf-pilot decision engine
nf-pilot makes the split-second choices: which of the 2,153 nf-core modules a step needs, where a request should go, and why a task died. It runs on the CPU in 175 ms, so the generative model is only called once the choice is already narrowed.
- 02
Deterministic build
A fixed sequence (extract, scaffold, install modules, stitch, lint, test, report) owns module names, include paths and configuration; the model drafts only subworkflow bodies, and each draft must pass validation.
- 03
nf-core scaffolding
Projects are created with the nf-core tooling and ship with a generated test profile, pinned container config and a fixture samplesheet, so a test run works from the first commit.
- 04
Container resolution
A local registry of container recipes and pinned composite images resolves single and piped multi-tool processes into a container config automatically.
- 05
Lint and autofix
An AST-based compliance suite plus native nf-core lint feed a fix loop that iterates until clean, with backups for rollback.
- 06
Migration and conversion
Rewrites DSL1 pipelines to DSL2 and converts Snakemake rules and WDL tasks, each with a record of what changed.
- 07
Test data on demand
Subsampled offline fixtures, in-silico read simulators, or streamed SRA and ENA samples, so pipelines are tested without staging large datasets.
02 / Questions
Often asked.
- Does the model write my Nextflow code?
- Mostly no. Drafts are compile-checked, nf-core linted and channel-checked before anything lands. A failing draft is never written to disk.
- Which models can it use?
- It is model-agnostic: hosted providers or local models through Ollama, with your own keys, so it can also run fully offline.
- How is a generated pipeline tested?
- Using the generated test profile and fixture samplesheet, Codaris runs a compile check, then a full local run against subsampled or streamed data, and retries with more memory on out-of-memory exits.
Early access · Codaris
Pilot Codaris on your Nextflow pipelines.
We are onboarding bioinformaticians and core facilities who maintain complex nf-core workflows. Run it locally, against your own models.
Request early access →