Today we're launching Phaide AI on Product Hunt. If you've ever suspected the answer you needed was sitting in your data behind a question nobody thought to ask — this launch is for you.
→ Support us on Product Hunt: producthunt.com/products/phaide-ai → Or go straight to the product: beta.phaide.ai/signup — free tier, 3 users, no credit card.
Why we built Phaide
Before Phaide, our team built two open-source projects: quelmap, a fully local data analysis agent, and Lightning-4b, a 4B-parameter LLM fine-tuned for data analysis. Conversations with the people who used them kept surfacing the same two problems:
- Most teams don't know what to ask their data. Chat-with-your-data tools still make you supply the right question. The insight teams needed was usually already in the data — behind a question nobody thought to ask.
- Many companies can't hand confidential data to an AI at all. Especially here in Japan, "AI analysis" was over before it began, because the data could never leave the building in readable form.
Phaide AI is our answer to both: an AI-native BI tool where the agents do the exploring — and where sensitive data is masked before any model ever sees it.
What's live today
Everything below is in the product right now at beta.phaide.ai:
- Autonomous exploration. Connect your data and Phaide's agents explore it on their own, then bring you the findings — flagging the genuinely surprising discoveries. No prompts. Discover the problems.
- The hypothesis-validation agent tree. Exploration runs as a tree: each node is an independent agent with its own hypothesis and its own Python sandbox. It tests the idea against your data, then branches into the next. Wide, parallel exploration surfaces patterns a single chat pass would miss — and every answer is computed by real code, not guessed.
- Automatic data masking. At ingest, confidential values — company names, customer records, internal IDs — are hashed to anonymized tokens before any model sees them. The mapping stays on your org's side. AI analysis without handing AI your data.
- Dashboards. The agent builds live dashboards from its findings, and you edit them conversationally. Share them with your team, with owner/editor/viewer roles.
- Documents, generated. Markdown reports with embedded charts, plus PowerPoint, Excel, and Word exports — analysis that arrives ready for tomorrow's meeting. Schedule explorations and route results to Slack or email.
- MCP + API. A REST API, API keys, and an MCP server, so your own agents and tools can drive Phaide programmatically.
Connectors cover CSV/Excel uploads, Postgres, MySQL, Supabase, BigQuery, Snowflake, ClickHouse, Databricks, Oracle, Redshift, and REST APIs — and the agent reasons across multiple sources in one place.
Who it's for
Teams that have data but no time to interrogate it: founders and ops leads who want problems flagged before they become fires, analysts who'd rather review twenty tested hypotheses than hand-write one, and security-conscious companies that could never use AI analysis before because of confidentiality. The product and site are fully bilingual (English / 日本語).
Pricing
- Free — $0, 3 users, 400 AI credits/mo, no credit card.
- Pro — $35/editor/mo, up to 10 users, 1,500 credits.
- Team — $99/editor/mo, unlimited users, 4,200 credits, MCP + Developer API.
- Enterprise — custom: SSO, self-hosting, custom connectors.
Viewers are free on all paid plans.
Come say hi
We're on Product Hunt all day today — the founder is in the comments, and honest feedback (especially critical feedback) is exactly what we're there for.
→ Support us on Product Hunt → Try Phaide on your own data — the discovery arrives before you ask.