Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.vlm.run/llms.txt

Use this file to discover all available pages before exploring further.

The vlmrun execute command submits agent executions via the /v1/agent/execute endpoint. Upload files, attach skills, and get structured JSON output.

Basic Usage

# Execute a named agent with a file
vlmrun execute -n my-agent:v1 -i invoice.pdf

# Execute with a prompt and schema
vlmrun execute -p "Extract invoice fields" -i doc.pdf --schema schema.json

# Execute with multiple files and toolsets
vlmrun execute -n my-agent:v1 -i a.jpg -i b.pdf -t image -t document

Agent Names

Use <agent-name>:<agent-version> format. If --name is omitted, the prompt is used to identify the agent.
vlmrun execute -n invoice-extractor:v2 -i invoice.pdf
vlmrun execute -p "Describe this image" -i photo.jpg

Using Skills

Attach skills inline from a local directory or reference server-side skills by ID. Only one of --skill or --skill-id may be provided.
# Inline skill from a local directory
vlmrun execute -n my-agent:v1 -i img.jpg --skill ./my-skill

# Server-side skill reference
vlmrun execute -n my-agent:v1 -i img.jpg --skill-id my-skill:latest

Models

ModelDescription
vlmrun-orion-1:liteLightweight
vlmrun-orion-1:fastSpeed-optimized
vlmrun-orion-1:autoAuto-select (default)
vlmrun-orion-1:proMost capable
vlmrun execute -n my-agent:v1 -i photo.jpg -m vlmrun-orion-1:pro

Toolsets

Enable specific tool categories with --toolset (repeatable):
vlmrun execute -n my-agent:v1 -i photo.jpg -t image -t web
Available toolsets: core, image, image-gen, world-gen, viz, document, video, web.

Async Execution

By default the CLI waits for execution to complete. Use --no-wait to submit and return immediately:
# Submit and return immediately
vlmrun execute -n my-agent:v1 -i photo.jpg --no-wait

# Check status later
vlmrun executions get <execution-id>

Output Formats

# Rich formatted output (default)
vlmrun execute -n my-agent:v1 -i invoice.pdf

# JSON output for scripting
vlmrun execute -n my-agent:v1 -i invoice.pdf -f json

# Pipe JSON to jq
vlmrun execute -n my-agent:v1 -i invoice.pdf -f json | jq '.response'

Prompt Sources

The prompt can be text or a path to a .txt/.md file:
# Inline text
vlmrun execute -p "Extract all line items" -i invoice.pdf

# From a file
vlmrun execute -p instructions.md -i invoice.pdf

Options Reference

OptionShortDescription
--name-nAgent name as <name>:<version>
--prompt-pPrompt text or path to a .txt/.md file
--input-iInput file (repeatable)
--schemaPath to a JSON schema file for the response model
--skill-kPath to a local skill directory (repeatable)
--skill-idServer-side skill as <name>:<version> (repeatable)
--toolset-tTool category to enable (repeatable)
--model-mModel variant (default: vlmrun-orion-1:auto)
--wait / --no-waitWait for completion (default: --wait)
--timeoutTimeout in seconds when waiting (default: 300)
--poll-intervalSeconds between status checks (default: 5)
--callback-urlWebhook URL called on completion
--format-fOutput format (json)