Frequently asked questions
Product positioning
Q: What's the fundamental difference between Deep Insight Agent and a meeting-secretary AI?
A: A meeting secretary "summarizes" — compressing and restating what others said. Deep Insight Agent makes "judgments" — pointing out whose words you can't trust, which consensus isn't real, and which silent signals were missed. A meeting secretary saves you reading time; Deep Insight Agent saves you judgment time.
Q: I already use ChatGPT for analysis — why do I need Deep Insight Agent?
A: General-purpose ChatGPT has no method library trained specifically for "meeting insight." Deep Insight Agent's 12 methods (Probe the Bones, First Principles, McKinsey, business-negotiation game theory) were iterated on hundreds of real founder and consultant recordings. One example: general AI won't proactively check the gap between "the consensus in the room" and "the real consensus" — Deep Insight Agent will.
Q: Who is Deep Insight Agent for?
A: Three core user groups: ① company founders (strategy meetings, funding negotiations, quarterly retros) ② independent consultants (client interviews, deep partner conversations) ③ corporate strategy teams / VCs (executive meetings, industry research, portfolio interviews). What they share: frequently holding high-density judgment meetings.
Data and privacy
Q: Will my recordings be used for training?
A: No. Deep Insight Agent does not use any user data to train LLMs. We use the Anthropic Claude API, and Anthropic commits to deleting API data after 30 days and never using it for training.
Q: Can other users see my analyses?
A: No. The database layer uses Supabase Row Level Security for hard isolation, so there's zero data crossover across organizations or users. We also have no access to your specific analysis content (the service role is used only briefly while the worker runs).
Q: Can it be deployed on-prem?
A: Yes. The Enterprise plan supports fully on-prem deployment: source code license, on-prem LLMs (Qwen/DeepSeek), and data that never leaves your servers. Contact enterprise@zaowuyun.com.
Q: How long is the original audio kept?
A: It's deleted automatically after 30 days (run by a cron job). For longer retention, the Enterprise edition is configurable; on the personal edition, you can download a backup manually.
How to use it
Q: Do I need to transcribe first?
A: Both ways work: ① upload an audio file directly and Deep Insight Agent runs Tongyi Tingwu for transcription + speaker separation ② paste an existing transcript (exports from Feishu Minutes / Tongyi Tingwu / Otter all work).
Q: How long does one analysis take?
A: Quick (under 5K characters) ≈ 30 seconds; Standard (20K characters + 1–2 methods) ≈ 1–3 minutes; Deep (full spectrum or a multi-method combo) ≈ 3–10 minutes. Tasks over 5 minutes go through the async queue.
Q: How do I know the analysis is reliable?
A: Every judgment is tagged [#claim-N] — click it to jump to the source segment in the recording. "Click-through to evidence" is Deep Insight Agent's core promise: a judgment with no source backing it = it doesn't exist.
Q: Can multiple people collaborate?
A: On the personal edition, each account is independent. The Team edition (from ¥499/seat/mo) supports a team memory base + shared analyses across members.
Pricing and credits
Q: How are credits calculated?
A: 1 credit ≈ ¥0.30 of real LLM cost. A "5K characters + single method" run ≈ 1–2 credits; "20K characters + Recipe A's four-method combo" ≈ 8–15 credits. Sign up for 100 free credits — plenty to put it through its paces (around a dozen analyses).
Q: Why are fewer credits deducted than estimated?
A: Prompt Caching saves a lot of cost: the second time you run the same method, SKILL.md + method.md come from cache, cutting cost by 60–80%. We bill by actual token usage, not by the estimate.
Q: Are credits charged if it fails?
A: No. Failures are refunded automatically (a two-way settle/refund ledger you can verify in your /settings transactions).
Q: Can I request an invoice?
A: Yes. After topping up, email hi@zaowuyun.com with your company name and tax ID to request one.
Methodology
Q: Are the 12 methods open source?
A: Yes. https://github.com/qiuyiwu1989-star/Deep-Insight-Agent, MIT licensed. Deep Insight Agent's commercial value lies in the memory system, collaboration, enterprise deployment, and the hardware channel — open-sourcing the methodology helps us earn community contributions and enterprise PoC validation.
Q: Can I contribute a new method?
A: Welcome. The open-source repo's CONTRIBUTING.md has a guide. Accepted new methods are synced into the product's method library, and contributors receive 100 bonus credits.
Q: How were the methods developed?
A: Based on hundreds of real meeting recordings by Deep Insight Agent's author, Qiu Yiwu, at Z+ Startup Camp, Lüedong AI, and several startup teams — iterated over 18 months. Every method has been validated across ≥30 real-world scenarios.
Hardware and the future
Q: Will Deep Insight Agent make hardware?
A: Yes, planned for 2027. Possible forms: ① a personal AI voice recorder (carried by consultants) ② a meeting-room box (for strategy teams) ③ a one-tap minutes button. Team or Enterprise subscribers get priority for the Beta.
Q: Is the API available now?
A: Yes. Create an API Key at /settings/api-keys and refer to the /docs/api documentation. It currently supports transcript push + analysis read; triggering analyses requires W6 (the org's shared credit pool).