Your research team, on demand

Deep research that shows its work.

Ask a hard question. Get back a cited report, checked against its sources, that you can hand to a client.

Free to start · no credit card · export everything

GLP-1 supply chain

GLP-1 supply chain: capacity, constraints, and the 2027 outlook

Complete9,412 words142 sources89 citations
verified

Executive finding

Fill-finish capacity—not active pharmaceutical ingredient production—is likely to remain the binding constraint through 2026. Announced expansions begin to relieve pressure in late 2026, but execution risk stays concentrated across eleven major sites.sec.govfda.gov

Projected branded dose capacity
202120242027E

Capacity is arriving in waves

The first meaningful relief comes from brownfield expansions, followed by new U.S. fill-finish lines.

Demand remains the larger uncertainty

Obesity adoption, reimbursement, and oral entrants widen the 2027 scenario range.

Workspace saved · every citation linked to a source file
01How it works

From hard question to defensible answer.

01

Ask

Describe the question in plain language. Attach files or connect your tools if they matter.

02

Approve the plan

It drafts a research plan first. Edit it, cut it, or run it as written.

03

Get the report

Minutes to an hour later: a cited report, checked against its sources, ready to export.

The plan

Nothing runs until you approve it.

Your question becomes a plan you can read in thirty seconds and rewrite in another thirty.
runs / glp-1-supply-chain-2027 / plandraft · editable

Map the GLP-1 supply chain through 2027

Identify the binding capacity constraint, model announced expansions against prescription demand, and name the companies most exposed if supply stays tight.

  1. 1.Define the market and demand baseline

    Establish branded and compounded volumes, prescription growth, and geographic scope.

  2. 2.Map manufacturing capacity

    Separate API, device assembly, and fill-finish constraints by company and site.

  3. 3.Model the 2027 supply balance

    Compare commissioning schedules against demand scenarios and flag execution risk.

  4. 4.Identify the exposed companies

    Rank manufacturers and downstream buyers by sensitivity to a sustained shortfall.

  5. Add section
  6. End of planRun
02Evidence

The failure mode of AI research isn't ignorance. It's confidence.

So every report is built to be checked: by a second model before you read it, and by you, claim by claim, after.

Verification

Every report is checked before you read it.

A second model re-reads the finished report against its saved sources — repairing what it can prove wrong, labelling what it can't.

Runs automatically on every research run

runs / glp-1-supply-chain-2027 / verification4m 12s

Audit passed with caveats

89 citations · all resolve to archived sources · 1 repaired · 1 caveat

  • Citation supportpassed

    Every claim resolves to a source saved with the run.

  • Analytical accuracyrepaired

    Draft read “Novo holds 70% of capacity.” The filing says “roughly half.” Corrected.

    checked against sources/sec.gov/novo-nordisk-20f-2025.pdf

  • Contradictionscaveat

    Two sources disagree on the 2026 commissioning date. Both cited; the report says so.

    checked against sources/sec.gov/lly-10k-2025.pdf

Provenance

Citations that can't rot.

Every source is saved into the workspace as a file, and every claim cites that copy. When the link dies, your evidence doesn't.

report · sources · data · charts → .zip

runs / glp-1-supply-chain-2027 / report.md142 sources saved

In the report

Fill-finish capacity — not active pharmaceutical ingredient production — is likely to remain the binding constraint through 2026sec.gov

resolves to a file saved in this workspace
sources/sec.gov/lly-10k-2025.pdf
“…fill/finish capacity remains the primary constraint on our ability to meet demand for incretin-based medicines…”

archived with the run · page 41

And everything else the run produced

  • report.md
  • sources/
  • sec.gov/lly-10k-2025.pdf
  • fda.gov/shortage-status.md
  • datafeeds/capacity-by-site.csv
  • visualizations/supply-demand.chart.json

Export .zip — report, sources, data, and charts

Primary sources

18 public-data feeds, including SEC EDGAR, FRED, the Federal Register, and Congress.gov, each with a specialist subagent.

Your tools

150+ connectors over OAuth, including your mail, docs, tickets, and CRM.

Open web

Multi-engine search, with allow and block lists per run.

See every source and connector
04Pricing

Start free. Upgrade when you need more.

Free
$0

For your first hard questions.

  • Included research allowance
  • All AI models
  • All connectors and primary sources
  • Cited reports, verified automatically
  • Full workspace export
  • 2 runs at once
Start free
Most Popular
Pro
Loading price

For people who research for a living.

  • Everything in Free
  • Expanded research allowance
  • 10 runs at once
  • Priority support
Start with Pro
Enterprise
Custom

For teams with stakes.

  • Everything in Pro
  • Custom connectors
  • Security and deployment review
  • Custom SLA
  • Custom support plan
Contact us

Current allowances and live checkout details are shown on the pricing page.

05FAQ

Fair questions.

How is this different from deep research in ChatGPT or Perplexity?

They're good, and often included in a plan you already pay for. The difference is what you're left with: a saved copy of every source the report cites, a check of the report against those copies, and the whole workspace to export. See the full comparison.

What do I get at the end of a run?

A cited report and the workspace behind it: every saved source, data file, and chart. Ask follow-ups in context, or export the whole thing as one archive.

How long does a run take?

A few minutes for a focused question, up to an hour for one that spans hundreds of documents. You can close the tab while it works.

Can it forecast?

Yes. Forecast mode turns a question about the future into a written resolution rule, has up to five forecasters research it independently, and scores the result when the date arrives. Make a forecast.

Do I need to know how to code?

No. You ask in plain language and approve a plan in plain language. If you do code, the same engine is available over a REST API, and runs can be scheduled or triggered.

Put your next hard question to work.

Free to start. The first report makes the argument for us.