The AI assistant that never guesses

Ask anything.
Get proof — or the right person.

One knowledge layer answers your customers, your team and their AI agents, and attaches the source every time. When it has no source, it does not guess: it hands the question to your people.

Free plan, no credit card Thai and English PDPA controls built in
Try:

This is the live assistant, answering from Puccha’s own documentation. Ask it something that is not there, and it will tell you.

9
chat channels, one inbox
14
industry assistant templates
2
languages, answered in the one asked
0
of your data used to train models

How It Works

From Documents to Cited Answers

01

Add your knowledge

Upload PDF (scanned pages included), Word, Markdown, CSV, HTML or audio files — or point Puccha at your website, which it re-crawls daily.

02

Connect your channels

One script tag puts the widget on your site; LINE, Messenger, WhatsApp and the rest connect from settings. People ask by text, voice note or photo.

03

Answer, cite, hand off

Each answer carries [1][2] links to its source. What your documents do not cover is declined and can go straight to your team.

Why Puccha

Answers you can stand behind

www.hospital.example — Puccha widget
Every answer shows its source
Real screens from Puccha, photographed on a fictional demo hospital.

Every answer shows its source

Puccha answers only from your documents and links the passage it used. When your documents do not cover a question, it declines and hands it to your team.

app.puccha.ai/knowledge-gaps
It gets better every week
Real screens from Puccha, photographed on a fictional demo hospital.

It gets better every week

Questions it could not answer become knowledge gaps, ranked by how many people asked. Approve one draft and every channel answers it from then on.

app.puccha.ai/conversations
One knowledge base, every channel
Real screens from Puccha, photographed on a fictional demo hospital.

One knowledge base, every channel

Your website, LINE, Messenger, WhatsApp and email draw on the same documents. Hand-offs land in one shared inbox for your team.

Start Free

Compliance

Controls you can show an auditor

PDPA controls operate today, and BOT, SEC and OIC obligations are mapped control by control in the trust centre.

  • Per-document access

    Each answer uses only passages the asker is allowed to see.

  • Retention you set

    Choose how long conversations are kept. Erasure on request.

  • Everything on the record

    Admin actions and staff access to conversations are audit-logged.

  • Any AI model, same guardrails

    Access control, citation checks and refusals are Puccha’s, so the model can change without changing your policies.

Security and compliance details →

Pricing

Simple, Transparent Pricing

Start free. Upgrade when you're ready.

Free

$0 /mo

100 resolutions/mo (hard cap)

Start Free
Most Popular

Team

$109 /mo

1,000 resolutions/mo

Start Free

Business

$549 /mo

5,000 resolutions/mo

Start Free

Enterprise

Custom

25,000+ resolutions/mo

Contact Sales

See detailed pricing & FAQ →

FAQ

Questions & Answers

What is Puccha?

Puccha (ปุจฉา — "question" in Pali) is the AI that answers with proof, or hands you to a human: one governed knowledge layer that answers your customers, your staff and their AI tools from your own documents. You add policies, manuals and web pages; Puccha answers customers and staff on your website, LINE and other chat channels with a citation for every answer, hands the rest to your team in a shared inbox, and shows you which questions your documents do not yet cover.

Is my data secure?

Every query is scoped to your organization, and cross-tenant access is tested on every endpoint. Documents can be restricted to a role or named users, and the restriction applies before the model sees a passage. Data is stored on Cloudflare; only the passages needed to answer go to the model and reranking providers listed in the trust centre. Your data is never used to train models.

Which AI model does Puccha use? Are we locked in?

Neither you nor Puccha is tied to one vendor. Answer generation runs through Cloudflare AI Gateway — today on Anthropic’s Claude, with a standby route — while embeddings and voice transcription run on Workers AI and reranking on Cohere. Every guardrail (retrieval grounding, per-document access control, citation validation, PDPA controls, answer-quality evaluation) is enforced by Puccha outside the model, so the model can change without your policies changing. Enterprise customers can discuss model requirements with us; data residency outside the current sub-processor regions is not offered today.

What happens when the AI can't answer?

It tells the person it does not have that information instead of guessing. Depending on your plan and settings, the conversation goes to an agent in the shared inbox, or a ticket is opened with an email notification. If the assistant itself fails, the customer is told and a person is brought in regardless of plan. The question also appears in your knowledge-gap queue, so the missing document can be written.

Can I cancel anytime?

Yes. Cancellation takes effect at the end of your billing period, and your organization moves to the Free plan — your documents and history stay available. To have everything erased, an owner can request erasure; it completes after a 7-day grace window.

More questions →

Put Your Documents to Work

Start on the Free plan. No credit card required.

Get Started with Puccha