Pica CLI · Agent-Native Infrastructure

Equip your agent.

Install pica. Your agent ships images, video, audio, and skill-driven workflows from any CLI runtime, against the same registry that powers Picadabra Studio.

View source

Transcript · pica@picadabrav0.2.15

Three commands to ship

Install once. The agent does the rest.

Install once. Bun is provisioned automatically when missing.
Your agent learns the Picadabra workflow as a local skill.
The agent ships. Outputs land in ConvexFS, ready to share.

What pica does

Every capability is one command.

Pica is opinionated about the boring parts: auth, asset pipeline, model registry, task lifecycle. Your agent gets the verbs.

pica generate: Submit any registered model. Inline @file references upload to ConvexFS automatically; the agent never touches a Provider API directly.
pica skill install / publish: Pull workflows from the public marketplace, or ship your own with one command. Skills include reference docs, hero assets, and a tested execution recipe.
pica model search / info: Discover models by capability, family, or feature. No hardcoded IDs in your prompts.
pica task wait / get / list / retry: Webhooks update Convex; the CLI polls for results. Timeouts adapt to image, audio, and video by default.
pica project switch / create: Scope work into isolated projects, each with its own assets and history.
pica prompt find: Search the curated Snap Prompt pool for proven creative seeds before starting.
pica assets list / get: Manage uploads and outputs through the ConvexFS-backed asset pipeline. Stream by default.

Ready to install

Skills your agent installs in one line.

Browse all skills

Video · Motion transfer
motion-control
Transfer motion from a reference video to any character image. Dance moves, gestures, poses.
Video · Talking head
kling-avatar
Lip-synced avatars from a single portrait and a script.
Image · Editorial
article-illustrator
Auto-illustrate long-form articles. A Type × Style × Palette system keeps a whole publication coherent.