CLI reference

The knowledge command-line tool creates, updates, and removes OKF bundles. Every command accepts a shared --model flag; some commands also accept command-specific options.

Synopsis

knowledge [--model <model>] <command> [options]

Global options

Option	Description	Default
--model <model>	A litellm-compatible model identifier. The provider is selected from the model string.	gpt-4o

The --model flag is accepted by every subcommand for consistency, but knowledge remove does not invoke the LLM.

Commands

create

Extract concepts from a source and write a new OKF bundle to a directory.

knowledge create [--validate] <input> <output>

Argument	Description
input	URL (http:// or https://) or local file path.
output	Directory to write the bundle into. Created if it does not exist.

Option	Description
--validate	Run BundleSerializer.validate() after writing and report any structural issues.

Example

knowledge create https://google.github.io/styleguide/pyguide.html ./pyguide

Exit codes

Code	Meaning
0	Bundle written successfully.
1	Source could not be fetched or read.

update

Re-extract concepts from a source and overwrite an existing bundle.

knowledge update <input> <bundle_dir>

Destructive.

update performs a complete replacement of the bundle directory. Concepts present in the old bundle but missing from the new extraction are removed. If you have hand-edited any concept file, back it up before running update.

Example

knowledge update https://google.github.io/styleguide/pyguide.html ./pyguide

remove

Delete one or more concepts from an existing bundle by ID.

knowledge remove <concept_id>... <bundle_dir>

Argument	Description
concept_id	One or more concept IDs to remove. Repeatable.
bundle_dir	Bundle directory to modify.

Example

knowledge remove outdated-section deprecated-topic ./pyguide

Exit codes

Code	Meaning
0	Every requested concept ID was found and removed.
2	At least one requested concept ID was not in the bundle. The CLI prints a warning listing every missing ID.
1	The bundle directory could not be read or written.

Environment variables

Variable	Purpose	Default
OPENAI_API_KEY	OpenAI authentication	—
ANTHROPIC_API_KEY	Anthropic authentication	—
OLLAMA_HOST	Ollama base URL	http://localhost:11434
KNOWLEDGE_MAX_BODY_SIZE	Maximum HTTP response body in bytes	52428800 (50 MiB)
KNOWLEDGE_REQUEST_TIMEOUT	HTTP request timeout in seconds	30
KNOWLEDGE_MAX_RETRIES	Retries for transient failures	3

See Configuration for the full list.

Exit code summary

Code	Meaning
0	Success.
1	Unrecoverable error (fetch failure, write failure, etc.).
2	Partial success: at least one requested concept ID was unknown.

Logging

The CLI writes log output to stderr at INFO level using the %(message)s format. This keeps stdout clean for future pipe-friendly output. Pass --quiet (planned) to suppress progress messages in scripts.