Skip to main content

​
Ravi Artifacts

ravi artifacts is Ravi’s generic artifact ledger. It stores durable outputs and their context so agents can inspect, reuse, edit, and explain artifacts later.

​
What It Stores

Each artifact can carry:
  • identity: id, optional semantic kind, title, summary, status, tags
  • file data: original path, blob-store copy, MIME type, size, sha256
  • generation context: provider, model, prompt, command, duration, tokens, and known cost when explicitly available
  • Ravi context: session, agent, task, run, turn, channel, account, chat, message
  • structured data: metadata, metrics, lineage, input, output
  • relationships: links to tasks, sessions, messages, projects, or arbitrary targets
  • audit events: created, updated, attached, archived
  • immutable versions: content snapshots with manifest data and version assets

​
Commands

All commands support typed --json output where applicable.

​
Versioning

Artifacts keep the mutable high-level record for current state, while versions snapshot content. Creating an artifact with a file, blob, URI, or structured output creates an initial version. Updating file/blob/URI/output content creates the next version. Metadata-only edits do not create a version unless you call ravi artifacts snapshot. Restoring a previous version reapplies that version to the current artifact and creates a new version for the restore, preserving the full history.

​
Image Integration

ravi image generate automatically creates one kind=image artifact per generated image. The image artifact records the generated file, blob copy, hash, provider/model, prompt, duration, usage/tokens when returned by the provider, and current Ravi session/channel context when available.

​
Provider Rule

There is no automatic provider fallback. If the selected provider fails, the command fails. To try another provider, retry explicitly with another --provider.