Accumulate structured data - arg.ai

A frequent shape: each run of a local job produces a batch of structured items, and you want them to accumulate in one place — deduped, queryable, editable by the team. The workspace handles this when you pick a structured file format and let the agent (or your script) append-edit it. The flow:

LOCAL JOB → batch.json → arg upload → CLAUDE/SCRIPT MERGES → /data/main.csv | .kanban

Concrete things that fit this pattern

Lead scraping. Pull commenters / followers / event attendees from a source → append to a CRM .kanban.
API polling. Pull new rows from a billing / CRM / analytics API → append to a .csv.
Log aggregation. Daily error sample from local services → append to an incident .csv.
Inbox digest. Read your local mail / Slack / Linear → append a row per noteworthy item.
Repo audit. Scan a local checkout for TODOs / FIXMEs / outdated deps → append per repo.

A worked example: scraped commenters → CRM kanban

# 1. Local scrape (anything that produces a flat JSON)
node ./scrape-commenters.js https://... > /tmp/batch.json

# 2. Hand the batch to arg
arg upload /tmp/batch.json --to /crm/inbox/$(date +%s).json

Then a Claude Code session connected to the arg MCP runs:

Read every JSON in /crm/inbox/. For each item, check whether its profile_url already exists in /crm/pipeline.kanban. If new, look up the profile via browse_url, judge fit against /crm/icp.md, and append a card to the inbox column with {name, profile_url, evidence}. Delete the batch file when done.

End state: /crm/pipeline.kanban is a draggable board in the arg UI, idempotent on profile_url, with full provenance back to the scrape batch. The local job changes (LinkedIn comments today, event attendees tomorrow, a webhook firehose next week). The workspace destination stays the same — and so do the team’s habits around it.

Next

To do the merge automatically the moment a batch lands — no agent session — see Automations & reactions (recipe 9, lead enrichment).