Agent context
Harvest models
4 SQL models that turn the raw Harvest tables into the reports people ask about. Deploy them in Weld and your agent queries the definitions behind every number.
How the set is built
- 8Raw tables
- 8Staging
- 4Core
- 4Analytics
Staging cleans each raw table once, core holds the logic, and analytics is the table dashboards and agents read.
- Invoice agingEvery sent invoice with what is still owed, how late it is and how long it took to get paid. Harvest's Outstanding Invoices report plus payment speed.grain: invoice · 1 raw table
- Project budgetBudget, spend and profitability per project, all time.grain: project · 7 raw tables
- Time reportHarvest's Time report: hours, billable hours, billable amount and cost.grain: day x user x project x task · 2 raw tables
- Weekly utilizationHours against capacity per person per week: Harvest's Team report and utilization numbers.grain: week x user · 3 raw tables
- Weld or dbt
- The same SQL in both dialects. Sync the folder with GitHub Sync, or copy the dbt models into your project.
- Check before you rely on it
- Reconcile a month against Harvest's own reports first. Each model's notes say what the synced data can and can't support.
Connect Harvest.
Your agents start with context.
Sync Harvest to your warehouse, deploy these models, and query them from Claude, Cursor or ChatGPT over MCP. Free for 14 days, no credit card.