enrich_analysis
ChatGPTAdd sections to a completed analysis this API started. Without modules, it finishes any core sections still missing: call it when get_analysis reports readyForExport: false, then poll again. A report created on the preuve.ai dashboard is refused with 403 REPORT_NOT_FROM_AGENT_RUN, and neither polling nor retrying changes that. Optional modules, for deep reports only: founderFit (how well the founder profile saved on preuve.ai matches the plan; this tool takes no profile, so a 400 INVALID_FOUNDER_PROFILE means the user completes theirs on preuve.ai first), proofOfDemand (public posts where people describe the problem, each quoted and linked to its source, with a suggested reply when one could be drafted), playbook (launch plan), trends (search-interest trend for the idea's keywords), community (public forum discussions about the problem). The analysis itself never collects trends or community, so those two exist only after this call. Each module generates once per report; a second request answers 409 MODULE_ALREADY_GENERATED, and per-module hints in the response explain each outcome. founderFit returns inline (up to ~90s); the others run in the background, so poll get_analysis. Personal reports only: Agency client reports are complete when status is COMPLETED and take no enrichment.
export_analysis
ChatGPTGet one COMPLETED analysis as structured JSON (schemaVersion 2): scores, verdict, risks, competitors, the market trend, and, for deep reports, researched market sizing plus full sections, pivots and citations. A starter scan returns market.tam/sam/som/growthRate as null (no research step stands behind them) with market.trend kept; do not fill the gap with a market size of your own. On a deep report, market.sizedFor names the market those figures cover (region, basis chosen, inferred or global, and a note); quote it with any TAM, SAM or SOM, above all when basis is inferred: the region was read from the idea text, not asked for. Takes the run id or report id of an analysis THIS API started; a report you created on the preuve.ai dashboard is not exportable here (get_analysis can poll it, and answers exportable: false). Relay the refusal text rather than picking a route: it names the way forward for that specific report. Starting a fresh analysis here always works. Call it once get_analysis reports readyForExport: true. Starter scans return some fields as null, with a top-level "locked" object naming them and a note explaining why. Relay that note instead of reporting the fields as missing; metric-level notes (projectionNote) resolve apparent figure mismatches the same way. For deep reports, request verbosity "summary" first unless the user needs all section text or raw evidence. A full deep report can exceed 90,000 characters, but this tool can return at most 65,000 compact JSON characters because it sends both text and structured output. Summary keeps every number and every short value (scores, ratings, labels), the closing verdict (sections.synthesis) and the Bottom Line narrative (details.verdict.narrative). It drops the other section prose and Trends data points. On the largest reports it then reduces raw evidence and how-to material to numbers and short text and shortens long text until the export fits that limit. It lists every change under details.omitted. If an export still exceeds the limit, use the report URL from get_analysis to read the full report. For an Agency client report pass workspace "agency" and its reportId: the export applies the same scope restrictions and sanitization as the dashboard, and client data is confidential. Do not publish or send it elsewhere without the user's instruction.
generate_ideas
ChatGPTGenerate startup ideas from interests and skills: each idea has a name, pitch, why-now, market saturation, audience and tags, grounded in live search. Returns { ideas, remaining, limit, pro, fit, fitReason, fitHint }. Some accounts have a daily cap of 5 generations per rolling 24 hours, counted down in the response limit and remaining; others have a ceiling of 20 per rolling 24 hours (15 with pro), shown as limit null and refused with 429 DAILY_LIMIT_REACHED. A failed generation is refunded. No scan quota is used either way. The Pro model (deeper ideas) runs by default on accounts that have it, and the response pro says whether the Pro tier was used (under load a Pro-tier run can be served by a faster fallback model); pass pro: false for the faster standard model. pro: true is not available on every account — when it is not, the call answers 403 instead of ideas, so omit it unless the user asked for the Pro model. Founder-profile tailoring is automatic whenever the account qualifies; fit: false opts out. When tailoring did not apply, relay the response fitHint rather than guessing why. Follow up the most promising idea with start_analysis (scanType "starter").
get_agency
ChatGPTRead the connected account's Consultant or Agency workspace in one call: the workspace, the caller's role, the shared project slots left (canRunScan is ELIGIBILITY, not live availability: it reads role, approval, workspace setup and slots left, never the service kill switch, the LLM pause or the rate limiters, so a start can still answer 503 or 429 with canRunScan true — neither uses a slot, so retry those later), and the client reports already in it — dashboard-created ones included — as { reportId, title, status, scanType, createdAt, reportUrl } without report bodies, newest first, with nextOffset for paging. Call it before starting client work so no shared project slot is used blind, and to answer "what did we run for this client" without starting anything. Then use start_analysis, get_analysis and export_analysis with workspace "agency". Agency membership is checked on every call; an account with no workspace gets 404 AGENCY_NOT_FOUND, and a key without the Agency scopes gets 403 INSUFFICIENT_SCOPE. Personal AppSumo Tiers 1-3 use the personal tools instead.
get_analysis
ChatGPTPoll one analysis by run id or report id to see whether it has finished. Any personal starter or deep report the connected account owns resolves, including reports the user created on the preuve.ai dashboard (the report id is the UUID in the report URL, /app/results?id=...); an Agency client report needs workspace "agency" instead. Returns status (PROCESSING/COMPLETED/FAILED), readyForExport, enrichment progress, and report/share URLs. A dashboard-created report answers with exportable: false plus exportBlockedCode REPORT_NOT_FROM_AGENT_RUN and never a pollAfterSeconds once it is COMPLETED: that is permanent, not a sequencing state, so do not call export_analysis or enrich_analysis on it and do not keep polling. While the run is in flight the response carries pollAfterSeconds; wait at least that long before polling again. Does NOT return the report body; use export_analysis for that. Note: an in-flight deep run reports analysisTier "basic" until deep sections land, so use scanType as the request-type signal. For an Agency client project pass workspace "agency" and the reportId from start_analysis or get_agency; those reports are complete when status is COMPLETED and need no enrich_analysis.
start_analysis
ChatGPTValidate a startup or business idea: runs a viability analysis across 60+ live data sources and returns a scored verdict with competitors, risks, and, on a deep scan, researched market sizing. Use it when someone asks whether an idea is worth building, whether a market exists, or who else is already doing it. Returns { id, reportId, scanType, status, ... }. Runs are async: a "starter" scan completes in about 1 minute, a "deep" scan takes at least 10 minutes. Poll with get_analysis until status is COMPLETED, pacing by the response pollAfterSeconds and doing other work in between rather than idling. scanType "deep" USES one of the account's deep scans, so default to "starter" unless the caller explicitly wants a deep scan. A PERSONAL deep scan also REQUIRES "stage" and "budget": ask the user for both BEFORE calling (never guess them) — the tool refuses to start a personal deep scan without them. To screen several ideas, call this once per idea and compare their exports, but pace the fan-out: one account holds 3 starter and 2 deep analyses in flight at a time and the next start answers 429 CONCURRENT_LIMIT_REACHED, which is pacing rather than failure. Wait for one to finish, then retry that idea with the SAME clientRunId. workspace "agency" starts a CLIENT project in the account's Consultant or Agency workspace instead: always deep, uses ONE of the workspace's shared project slots, never falls back to the account's own quota, returns { reportId, status, ... } and cannot publish a share link. The stage/budget requirement above does NOT apply there: a client project takes both as optional free text, so pass them only as the client stated them and start the project without them otherwise — never hold one waiting for context the consultant does not have. Use it only when the user asks for client work, after get_agency has shown the slots left. Reuse clientRunId after a timeout to recover the original run, so the same idea is not started twice; then poll get_analysis with the same workspace.
create_batch
Claudeenrich_analysis
Claudeexport_analysis
Claudeexport_batch
Claudegenerate_ideas
Claudeget_agency
Claudeget_analysis
Claudeget_batch
Claudestart_analysis
Claude