Artifacts

The event log stays lightweight because it never embeds file bytes. When inline media enters the conversation, the bytes are stored as an artifact (art_…) and the persisted event carries a media_ref block pointing at it:

{ "type": "media_ref", "artifact_id": "art_z3jehcyd5n6a2bfgik7mv4qtrw" }

In the dashboard, artifacts appear as downloadable cards or attachment chips on the conversation timeline; rendering your own timeline means resolving each media_ref with the two endpoints on this page. Model context identifies conversation attachments by their public artifact_id, allowing an agent with download_file to copy one to an attached machine using path: "/artifacts/<artifact_id>" and a machine destination.

Artifacts uploaded by an agent

A successful upload_file result contains the path as structured data and a media_ref for presentation. Use the artifact ID from the final component of the path with the metadata and content endpoints below. The relevant fields in the resulting event look like this:

{
  "event_kind": "tool_result",
  "tool_call_id": "tcl_wz3jehcyd5n6a2bfgik7mv4qtr",
  "outcome": "succeeded",
  "content_blocks": [
    {
      "type": "structured_data",
      "value": {
        "path": "/artifacts/art_z3jehcyd5n6a2bfgik7mv4qtrw"
      }
    },
    {
      "type": "media_ref",
      "artifact_id": "art_z3jehcyd5n6a2bfgik7mv4qtrw",
      "exclude_from_model_context": true
    }
  ]
}

export BASE="$OMNARA_API/orgs/$ORG/projects/$PROJ"
export AGENT="agt_v4qtrwz3jehcyd5n6a2bfgik7m"

Get artifact metadata

Fetch the descriptor first when you need to decide how to render or whether to download:

curl "$BASE/agents/$AGENT/artifacts/art_z3jehcyd5n6a2bfgik7mv4qtrw" \
  -H "Authorization: Bearer $OMNARA_TOKEN"

{
  "id": "art_z3jehcyd5n6a2bfgik7mv4qtrw",
  "org_id": "org_k7mv4qtrwz3jehcyd5n6a2bfgi",
  "project_id": "proj_7mv4qtrwz3jehcyd5n6a2bfgik",
  "agent_id": "agt_v4qtrwz3jehcyd5n6a2bfgik7m",
  "content_type": "application/pdf",
  "filename": "eu-ai-act-summary.pdf",
  "digest": "sha256:9f2ab8…",
  "size_bytes": 482133,
  "created_at": "2026-08-03T19:15:40Z"
}

digest is a content hash — useful for caching and deduplication on your side: if you’ve already stored that digest, skip the download.

Full schema and playground: Get artifact.

Download the content

The content endpoint returns the raw bytes with the artifact’s content type:

curl -o eu-ai-act-summary.pdf \
  "$BASE/agents/$AGENT/artifacts/art_z3jehcyd5n6a2bfgik7mv4qtrw/content" \
  -H "Authorization: Bearer $OMNARA_TOKEN"

Content is immutable — the same artifact ID always yields the same bytes, so responses are safe to cache indefinitely.

Full schema and playground: Get artifact content.

Where artifacts come from

Source	What happens
Inline `media` block on an input	Persisted as an artifact; the stored event shows a `media_ref`.
Inline `media` block on a custom tool result	Same conversion — submission accepts bytes, the timeline stores a reference.
`upload_file`	The durable tool result stores the path and a `media_ref`.
Model output with media	The `model_output` event references it.

There’s no general-purpose public endpoint for creating artifacts; they’re created during agent interactions.

Streaming events

Where media_ref blocks appear in each event kind

Sending input

Attach files to your messages

Get started

Examples

Agents

Inputs & events

Tools

Machines

Organization

Integrations

Artifacts uploaded by an agent

Get artifact metadata

Download the content

Where artifacts come from

Next

Streaming events

Sending input

​Artifacts uploaded by an agent

​Get artifact metadata

​Download the content

​Where artifacts come from

​Next

Streaming events

Sending input

Artifacts uploaded by an agent

Get artifact metadata

Download the content

Where artifacts come from

Next