Skip to content

The Interactive REPL

Codaris includes a rich interactive REPL for bioinformaticians and pipeline engineers. It lets you build, inspect, lint, and repair pipelines in a continuous conversation from the terminal.

The REPL is a thin terminal client. It streams every turn to the Codaris API at CODARIS_API_URL (default http://127.0.0.1:8000), so the API must be running, either through codaris web or through the unified Docker image. When the API is not reachable, the REPL prints a warning at startup and a connection error on the first turn.

Launching the REPL

Run codaris with no arguments:

codaris

Pass an opening prompt directly, and the REPL sends it as the first turn:

codaris Build a bulk RNA-seq pipeline for human paired-end FASTQ files

The explicit form accepts the model, workspace, and mode as options:

codaris chat --workspace ./my-pipeline --mode review

Launch from inside an existing pipeline directory to make it the active workspace:

cd ./my-pipeline
codaris

REPL Features

1. Natural Language Prompting

Describe the change you want in plain language:

codaris ❯ Add Picard MarkDuplicates after the BWA-MEM2 alignment step
          and wire the output BAM into GATK HaplotypeCaller.

2. Streaming Tool Traces

Each turn streams the agent's tool calls and their results as they happen. Toggle /verbose to see full tool call signatures and raw outputs when you are debugging a build.

3. Prompt History

Prompt history persists in ~/.codaris_history, so the Up and Down arrows recall prompts from earlier sessions. Chat context lives in the backend session for the current run; /clear starts a fresh session.

Slash Commands

Two kinds of slash command are available. Local commands are handled by the terminal client itself. Every other slash command is forwarded to the backend, where the same command router that serves the web studio handles it.

Local commands

Command Description
/help, /h, /? Show the help screen
/model [name] Show or change the current model
/mode [build\|plan\|review] Switch the agent execution mode
/workspace [path] Inspect or switch the active directory
/doctor Check toolchain and container health
/status Show session tokens, model, and active parameters
/verbose Toggle detailed tool arguments and debug tracing
/clear Clear chat history and start a fresh session
/exit, /quit, /q Exit the REPL

Forwarded commands

Command Description
/build [assay] Compile a Nextflow DSL2 pipeline into the workspace
/plan Produce a BuildPlan without compiling
/review Structural wiring review (channels, params, containers, configs); read-only, no scorecard
/audit Read-only compliance audit with a scorecard (/review --compliance is the same)
/remediate Apply rollback-safe compliance fixes
/fix Diagnose a Nextflow failure log and apply a targeted patch
/lint Run nf-core pipelines lint on the active pipeline
/rules List the compliance rule catalogue
/searchData Query NCBI SRA and ENA and write minimal FASTQ fixtures
/migrate Transpile Nextflow DSL1 to DSL2
/convert Convert CWL or WDL to Nextflow DSL2
/research Literature synthesis with PubMed and arXiv citations
/run Execute the active pipeline with the test profile

Keyboard Shortcuts

  • Up and Down: Browse prompt history.
  • Ctrl+C during a turn: Interrupt the current streaming generation and return to the prompt.
  • Ctrl+C at the prompt: Print a reminder; the session stays open.
  • Ctrl+D: Exit the REPL.