CLIO Coder

Check your installation

Read the full guide
On this pageStart with the standard check Read the findings Common rows and their fix HPC toolchain rows Repair or investigate further Related tutorials

Check your installation

Diagnose installation and connection problems with doctor.

Start with the standard check

Run doctor from your project:

Shell

clio-coder doctor

It checks the installation, configured connections, model-list evidence, local worker capacity, and relevant tools. Standard doctor is read-only, uses passive probes, and sends no model generation request. In the terminal, /doctor shows the findings in your session.

Read three rows first: chat names your route and what is missing, each connection <id> row says whether the key is there and the endpoint answered, and shell sandbox says whether commands run in an OS sandbox. Fix !! rows first; a WARN row still lets CLIO Coder work.

Flag Effect
--verbose Print every row instead of folding quiet families
--json Emit findings as JSON on stdout
--deep Add the live tool probe and validation dry run
--tools-timeout <seconds> Bound each tool probe; default 120, needs --deep

Doctor exits 0 when no row is an error, 1 when one is, and 2 on a usage error.

Read the findings

Badge Meaning
OK The check passed
INFO A fact that usually needs no action, such as an HPC compiler that is not on PATH
WARN Worth attention; CLIO Coder can still work
!! A broken requirement; doctor exits with an error

A live model list, a cached list, and a provider catalog are distinct evidence. A reachable endpoint does not establish that a model will answer well or call tools. The capacity row reports CPUs and available memory; it does not measure GPU memory or model fit.

Common rows and their fix

The first row, chat, says whether chat can run and what to run when it cannot.

Row Usual cause Fix
settings.yaml Invalid keys, named in the row Edit those keys
credentials credentials.yaml is missing or not 0600 clio-coder doctor --fix
lifecycle migrations Migrations are pending after an update clio-coder upgrade
connection <id> Missing credential or no endpoint answer Check the target and sign-in
model <id> The role model is missing from a live list Pick a listed model
external tool <id> A pinned helper is missing clio-coder tools install <id>

HPC toolchain rows

toolchain <name> rows cover compilers, MPI, nvcc, build tools, python3, and sbatch. An absent tool is INFO. It is WARN when your validation contract's validators need it, or when a tool's --version fails, as with an unconfigured Slurm client.

Repair or investigate further

clio-coder doctor --fix creates missing directories and template files, makes settings.yaml and credentials.yaml owner-only, rewrites retired setting values such as safety.autonomy: auto-edit and YAML on and off booleans while keeping your comments, and records fleet preflight results. Plain doctor previews the settings rewrite. Version migrations still run through clio-coder upgrade.

The shell sandbox row says whether this host contains your commands and where they may write.

For a live tool-use probe and validation-contract dry run:

Shell

clio-coder doctor --deep

Deep checks send model requests and may consume tokens. Allow more time for a cold local model with --tools-timeout <seconds>. Use --json for structured findings.

A tools <target> row is OK when the model streamed a valid tool call and WARN when the call was malformed or the target did not answer. A validator <n> row says whether each validator command would run without an approval ask; nothing is executed. In a session, use /doctor --deep.

If a connection remains unavailable, follow troubleshooting.