Quick Verdict
Tiangong AI should not remain cataloged as only Kunlun’s early answer-style search engine. On July 18, 2026, the official tiangong.cn title and application described a DeepResearch-based office super-agent with professional skills and generation workflows for documents, presentations, spreadsheets, and websites. Search remains an evidence-gathering layer, but the current buying decision must include artifact editing, export quality, credits, workspace permissions, and execution reliability.
Tiangong AI Music is a separate product boundary. music.tiangong.cn still presented a music service on the review date, but music generation, licensing, allowances, and payment terms must not be mixed into the office agent entry. Historical fixed membership prices, “completely free” claims, and early launch superlatives no longer describe the current product reliably.
Best For
- Chinese-speaking professionals who want public-web research to continue into a document, deck, table, or webpage draft.
- Analysts, educators, product teams, and operators prepared to verify citations and revise generated artifacts.
- Individuals or small teams piloting a Chinese research-to-deliverable agent workspace.
- Users who can test connected tools in a sandbox and approve any public, transactional, or data-changing action.
- It is not ideal for autonomous publication, exact complex financial spreadsheets, or users who treat the music and office subscriptions as interchangeable.
Key Features
- DeepResearch: expands a complex question into retrieval and a longer research output. It can locate evidence but cannot certify a source or conclusion.
- Office artifacts: the current entry emphasizes documents, presentations, spreadsheets, and websites created from research and prompts.
- Professional skills: specialized workflows can structure different tasks, but a skill name is not professional accreditation.
- Files and workspace: uploaded sources and persistent project context reduce repeated input while increasing sharing and retention responsibilities.
- Editing and export: generated artifacts can be refined. Layouts, formulas, references, download formats, and permissions require direct testing.
- Tool execution: website publishing, databases, application keys, external services, and payments, where available, should be tested with least privilege and human approval.
Use Cases
For an industry report, define geography, dates, terminology, primary sources, exclusions, and the expected structure before research begins. After DeepResearch, remove syndicated duplicates and verify policies, statistical definitions, company disclosures, and original studies. Mark unknowns rather than asking the model to fill them. For simpler Chinese long-document work, compare Kimi; for focused Chinese search, compare Metaso; for broader international source research, compare Perplexity.
For presentations, approve the factual narrative first, then inspect titles, chart axes, units, citations, and readability. For spreadsheets, recalculate formulas, data types, dates, missing values, and totals independently. A rendered table is not proof that the computation is right. Publish generated websites to a test location first and check mobile behavior, forms, access rules, external scripts, and secrets. Any order or payment path requires a responsible person to verify amount, recipient, callback, cancellation, and refund behavior.
Save source files, prompts, and intermediate versions outside the platform. Long agent tasks can fail after producing useful partial work, and an export should remain editable in ordinary office tools.
Pricing
The current application contains membership and credit-related concepts, but the public homepage did not provide one stable, comprehensive renminbi price table. This page removes the old ¥30/month statement and the contradictory claim that the service is entirely free. Buyers should inspect the live account for subscription period, task or model allowances, artifact exports, website publishing, file storage, failed-task refunds, renewal, tax, refund, and invoice terms.
| Area | Confirm before purchase | Pilot test |
|---|---|---|
| Free access | DeepResearch, file, artifact, and export limits | Complete one short research task and export |
| Membership/credits | Per-task use, reset, top-up, and failure treatment | Record retries and accepted-output cost |
| Documents/slides/sheets | Templates, page limits, formulas, formats, and watermarks | Use a complex representative file |
| Website/external tools | Publishing, domains, databases, keys, and integrations | Connect only a least-privilege test account |
| Team capabilities | Members, sharing, logs, data terms, and support | Create roles and verify visibility and revocation |
Pros
- Joins Chinese DeepResearch with several office deliverables in one entry.
- Research can flow into editable artifacts without repeated copy-and-paste.
- Chinese interaction and local account workflows are accessible to domestic knowledge workers.
- Professional skills offer structured starting points for distinct tasks.
- Suitable for a small pilot covering the full research-to-export path.
Cons
- Historical search-only reviews no longer represent the current super-agent product.
- Public pricing and credit allowances are incomplete and require account testing.
- Citations, reports, and charts can contain stale sources, unsupported claims, or calculation errors.
- Dedicated office, analytics, or web-development products may provide deeper control.
- The office and music products share a name but not automatically the same rights or plans.
Alternatives
| Tool | Best for | Relative strength | Main tradeoff |
|---|---|---|---|
| Genspark | International research and multiple agent artifacts | Broader international agent workspace | Access, credits, and permissions need testing |
| Kimi | Chinese long-document reading and writing | Straightforward long-context interaction | Office artifacts are less central |
| Metaso | Chinese search and structured reading | More concentrated search workflow | Fewer office and website deliverables |
| WPS AI | Editing inside established office formats | Deeper office-suite integration | Open-web DeepResearch is not the sole focus |
| Perplexity | Global source-led research | More focused citation workflow | Less localized office-artifact production |
FAQ
Is Tiangong AI still just a search engine?
No. Search and DeepResearch remain foundational, while the current official entry emphasizes an office super-agent producing documents, slides, sheets, and websites.
Is Tiangong AI Music included in this product?
Do not assume so. music.tiangong.cn is a separate product line. Accounts, credits, licensing, and prices should be verified independently.
Can a cited research report be submitted directly?
No. Open every material source, check author, date, original statement, statistical definition, and contrary evidence, then obtain domain review where consequences are significant.
Are generated slides, sheets, and websites ready to deliver?
Not by default. Check deck layout and figures, recalculate spreadsheet formulas, and test website behavior, permissions, forms, security, and integrations.
Is the old ¥30 monthly membership still current?
It cannot be treated as current. Use the signed-in checkout and entitlement pages, including credits, renewals, failed-task treatment, and refunds.
Can confidential company files be uploaded?
Only after organizational privacy, security, contract, retention, model-use, and permission review approves the data class. Start with public or de-identified material.
Bottom Line
Tiangong AI now makes sense as a Chinese research-to-office-artifact agent rather than an early search product. Re-evaluate it from the current entry: verify sources, test each artifact, record credit use, inspect exports, and constrain permissions. Keep the music line separate. A citation, attractive page, or completion indicator is never the final acceptance test; human review and an editable backup remain essential.