Build a Research Agent with AIsa

​

The problem

​

Prompt

Use AIsa to research: <your question>. Time range: <e.g. last 90 days>. Preferred sources: official docs, filings, original announcements, peer-reviewed papers.
Find the right AIsa search tools with `search`, and quote each paid call with `batch_quote` first. Ask me before any paid call, and tell me the expected number of calls.
Retrieve before you synthesize: search for candidates, then fetch the full text of every source you cite. Do not cite anything you did not retrieve.
Return a brief with: (1) verified facts, each with URL and publication date; (2) interpretation; (3) contradictions between sources; (4) open questions; (5) the research cutoff time.

​

APIs and Skills used

​

Example output

## Brief: Has Example Corp shipped its v2 API? (cutoff: 2026-01-15 09:00 UTC)

Verified facts
- Example Corp announced v2 general availability on 2026-01-10. [example.com/blog/v2-ga, 2026-01-10]
- The v1 endpoints are deprecated, not yet removed. [docs.example.com/changelog, 2026-01-11]

Interpretation
- Migration pressure is moderate: no removal date is published.

Contradictions
- A third-party post says v1 shuts down in March; the official changelog gives no date. [example.org/news/123]

Open questions
- Pricing for v2 is not stated in any retrieved source.

​

Guardrails

• Retrieve, then synthesize. Snippets and model recall are leads, not evidence. Fetch the full source for any material claim.
• Prefer primary sources. Prioritize primary sources (official docs, filings, original announcements, datasets) and record both the URL and publication date.
• Surface contradictions. Report conflicting information instead of merging it, and mark claims that rest on a single source.
• No invented citations. The model must not cite or claim access to sources that were not retrieved.
• Verify before delivery. Verify that each citation resolves, recheck time-sensitive numbers, remove unsupported claims, and state the research cutoff time.
• Quote, then ask. Get a free quote with batch_quote and run the same calls with batch_use only after approval. A quote checks the estimate, not the final charge. See Billing & Credits.
• Load only what you need. Ingest only the selected Skill or API documentation pages, Authentication, and Models. Do not load the entire API reference.
• When AIsa may not be necessary: A single specialized search provider can be simpler if you use only one source type, need no reusable Skills, and already manage your own model and billing integrations. See When to use AIsa.

​

Related