History

The table

Runs, newest first. Columns: Started, Type (scrape, crawl, research, extract, summarize…), Run (the run id), Status, Pages ok/err, Fetched (bytes), Tokens, Cost, and the fetch time. The desktop and the CLI share one store, so a script’s runs appear the moment they finish.

Filters

The filter bar composes:

• Type — All, Scrape, Crawl, Other.
• Status — Any, Finished, Failed, Running, Stopped.
• From / To — dates, YYYY-MM-DD.
• The text field matches a run id or a command substring.

Clear resets everything. The filters also arrive prefilled from Overview’s Activity tables, which deep-link into the matching slice.

The run drawer

Click a row. The drawer opens on a summary — status, timings, counters, and on a failure the reason bucketed as challenge, quality, timeout, robots, network or other, with the verbatim error. Three tabs:

• Output — what the desktop kept. A scrape: its Markdown and extracted record, read back from the replay file. A crawl: a preview of its records (the first 50, long cells truncated for display), the format picker with Export…, and Open folder for the run’s files on disk. Runs that keep no output say which: failed runs, CLI-started runs, watch checks and search hits were never saved.
• Reproduce — the run as a CLI command, an MCP tool call or a Go snippet, built from the options the run saved. Secret material stays redacted: a per-run key or pasted cookies keeps only a (redacted) placeholder. Runs with no saved options — the same list as above — say so instead of guessing.
• LLM calls — the per-call ledger behind the cost column: purpose, provider, model, prompt and completion tokens, dollars. A zero-extraction run shows none.

Resume and Re-run

A stopped or failed crawl keeps its checkpoint, and the drawer offers both continuations:

• Resume — continues the run under its own id. Done pages are not refetched; the new records rotate to a numbered part file (.part001.jsonl, .part002.jsonl, …) beside the original, and exports read every leg in order. A resume cannot change the mode, the schema or the seed — start a new crawl for that.
• Re-run — fills the Crawl form with the run’s saved options for a clean start under a new id.

The same two actions serve a research run from its drawer: the report, its evidence and Resume live on the Research screen.

Read-only mode — trial used, licence lapsed — keeps History fully working: reads, exports and copies are not gated.