Mavera DocsMave agent
Checking your Mavera sessionLoading workspace audiences and API keys…
Checking…Use or share this pageSend the page itself, print it, or open its complete context in your AI workspace.
Mave research agent
Web search + Python sandbox, grounded and cited.
Endpoint: POST /api/v1/tools/deep_research. Runs on the ORCHESTRATOR tier (Opus) in a multi-step agent loop.
Request
{ "question": "Fastest-growing SMB fintech segments in 2026, with sources.",
"context": "We're pricing a $29/mo finance app for solo founders.",
"web": true, "code": true }| Field | Meaning |
|---|---|
question | Required. The analytical question to answer. |
context | Optional background to focus the research. |
web | Allow live web search + fetch (default true). |
code | Allow the sandboxed Python interpreter (default true). |
Response
{ ok:true, data:{
answer: "…",
sources: ["weforum.org", "cbinsights.com", …],
toolTrace: [{ kind:"web_search", detail:"…" }, { kind:"code_execution", detail:"…" }],
ranTools: true, asOf: "2026-06-30",
honesty: { basis: "live-evidence", note: "…" } } }What it's good for — and not
- Good: current market data, competitor/pricing facts, computed comparisons, anything needing fresh external evidence with citations.
- Not: audience opinion.
deep_researchnever fabricates synthetic-respondent views — for that, field a study or run a conversation. Thehonesty.basisislive-evidence, distinct from synthetic-audience output.
Every figure is cited to a fetched source or computed in the sandbox. When the live agent is unavailable (no key/offline), it returns ranTools:false and an honest message rather than guessing.
Pair it with flip-tests
A natural loop: find mode hypothesizes a flip-condition → deep_research checks whether that condition is realistic in-market → flip-test verifies whether it actually moves the persona. Real-world plausibility + synthetic reaction, cleanly separated.