CLI Guide

AXIOM CLI Guide

Use the CLI for repeatable commands, scripts, and workflows that should not depend on clicking through the Control Center.

First commands

axiom version
axiom --help
axiom guide
axiom summary
axiom summary --json
axiom status
axiom doctor --json

--json output is intended for machine processing where supported. Human-oriented text can change; automation should prefer structured output and handle non-zero exit codes.

By task

Project and config

axiom init my-ai
axiom project validate
axiom config validate
axiom info

Models

axiom model list
axiom model inspect ./models/my-model
axiom model search llama --path models
axiom model info org/model
axiom model analyze org/model

Remote metadata and downloads need network access. Private or gated Hugging Face repositories may require axiom hf login. Model inspection does not itself run inference.

Datasets

axiom dataset inspect ./data/train.jsonl
axiom dataset validate ./data/train.jsonl
axiom dataset stats ./data/train.jsonl
axiom dataset profile ./data/train.jsonl
axiom dataset manifest ./data/train.jsonl
axiom dataset preview-clean ./data/train.jsonl

Dataset paths currently focus on JSONL. Validation checks structure; it does not establish semantic quality or guarantee that data is appropriate for a particular model. PII scanning is opt-in when supported and uses heuristics rather than a compliance guarantee. dataset clean is a separate explicit command that writes an output file; preview-clean does not write the cleaned dataset.

Hardware and planning

axiom system info
axiom train plan 7 --method auto
axiom train plan 7 --method lora
axiom train plan 7 --method qlora
axiom train plan 7 --method full
axiom ai plan --model ./models/my-model

Fit and VRAM values are planning estimates, not measured memory use, speed, or a promise that a later runtime will succeed.

Experiments and evaluation

axiom experiment list
axiom experiment show <experiment-id>
axiom evaluation run ./data/eval.jsonl --prediction-field prediction --reference-field reference

Experiment entries are plan snapshots with metadata, not evidence that a training run occurred. Evaluation scores precomputed predictions; it does not run inference.

Workflow design

axiom workflow list
axiom workflow show <workflow-id>
axiom workflow validate <workflow-id>

These commands inspect and validate saved designs. They do not execute workflow steps.

Diagnostics and support

axiom doctor
axiom doctor --json
axiom roadmap audit --json
axiom support-bundle --output ./axiom-support-bundle.zip

Read Workspace Diagnostics and Support Bundles before sharing archives.

Tools and MCP

axiom tools list
axiom tools list --json
axiom tools doctor
axiom mcp serve
axiom mcp permissions

Tool discovery does not prove that a vendor account is authenticated. MCP is a separate stdio integration and tools are denied by default; review the active policy and explicitly grant only the tools needed by your client.

Exit codes and automation

For scripts, use the command's documented JSON mode where available, check the process exit code, and avoid parsing human prose when a stable structured field is available. Do not assume every command family supports identical flags or output modes; confirm with axiom <command> --help on the installed version.

Related pages: Getting Started, Workspace Diagnostics and Support Bundles, Workflow Design and Execution Boundaries.