← Use cases

How do you know which agent or model produced this output?

A report shows up three weeks after the pipeline ran and something in it looks wrong. Before you can fix it, you need to know which agent produced it, which model it called, and which run it came from, and none of that is in the file itself.

Tag the push or publish call with an agent, a model, and a session, and that context travels with the artifact. On a published page, it surfaces as a provenance receipt behind a small info icon: agent, model, session, and when it was created, all in one place instead of scattered across logs.

If you don't pass an agent or model yourself, MCP clients stamp a default so the receipt is never empty. See the receipt on the same page a nightly job published on its own; the fields are what let you trace it back.

Try it

artifacta publish ./analysis.html \
--session pipeline_run_42 \
--agent research-agent \
--model claude-sonnet-5
 
artifacta inspect art_2xk9f7v3m1p0 --json | jq '{agent_id, session_id, metadata, created_at}'
{
"agent_id": "research-agent",
"session_id": "pipeline_run_42",
"metadata": {
"model": "claude-sonnet-5"
},
"created_at": "2026-07-04T03:12:00Z"
}

Live proof

See the provenance receipt on a live page
Get started for free

Related