Shard

See why AI work ran there

Flagship demo

Ask one question. Watch Shard show its work.

Ask one question, paste a few sources, and Shard will return the answer plus the receipt trail that explains where each step ran, what failed, and what the fallback cost.

1. Submit

Add a question and a small source bundle.

2. Route

Shard chooses between personal, private, and public capacity.

3. Explain

Receipts and the graph explain each decision in plain language.

Plain-language translation: provenance is just the step-by-step map of what happened during the run.

Workflow input

Try the research brief demo

Reload a stored execution

Open an older run by execution ID without rerunning the workflow.

Choose a starting policy

Start with a preset, then fine-tune the rules below if you want.

Execution policy

Tell Shard what it is allowed to use, how careful it should be, and how much it can spend on public capacity.

Allowed supply

These buttons decide which machines Shard may use at all.

Fallback order

This order decides where Shard looks next if the first choice fails.

Public spend cap

This is the most Shard may spend on public capacity for one run.

1. personal
2. private
3. public

Execution summary

Run the demo to inspect the workflow trail

This panel shows the finished answer first, then the map of how Shard got there.

Run the workflow to get the answer, the receipt chain, and the graph together in one place.

Step-by-step map

Why the workflow ran this way

Run the workflow to populate the receipt graph that Shard rebuilds from `parent_receipt_id` links.

Receipts

Raw receipts