dokusan.ai — A Conversational Database That Runs on Your Hardware

dokusan.ai

The database you always wanted

Your data texts you. In English. You text back. A conversation that results in insight.

dokusan ● your company

I found 14 customers who churned in Q3 despite having NPS scores above 90. The most common pattern is a pricing tier change in Q2 — 9 of the 14 had one.

what were the tickets about?

Two were about billing discrepancies after the tier change. The third was a feature request for bulk export that's been open since March.

what's the phone number for our HVAC vendor?

Comfort Air Systems — (925) 555-0142. Ask for Danny, he's the one who knows your units. Last service was January 14th.

who handled the Johnson account last time?

Maria Chen closed it in October. She noted they're price-sensitive and prefer net-30. The original intro came through the Riverside Chamber event.

what's our return policy on opened items?

Per your policy doc from March: opened items can be returned within 15 days with receipt for store credit only. Exception: defective items get a full refund within 30 days.

Heads up — that invoice from Meridian Supply you asked about in March just showed up in yesterday's email. $4,280. It's 47 days past the original due date. Want me to flag it for accounting?

Ask your data something…

Why dokusan

Data.

A traditional database is a library. You must learn the catalog system. You must know which shelf. You must speak its language.

Dokusan is the monk who already read every book and is sitting quietly, waiting for you to ask.

Technical details ↗

Privacy

Trust.

Dokusan keeps your data on your hardware — no third-party servers, no logs, nothing to discover. No open-source supply chain. No transitive vulnerabilities. Our code is ours and auditable.

Why this matters ↗

Get started

Hire dokusan.

Your newest employee knows everything, never sleeps, never quits, and answers from your phone at 2am. Your newest hire is already up to speed on day one.

Runs on your hardware — a Mac Mini under your desk, in our data center, or both. Your data never leaves your machine. Non-discoverable. Can't be subpoenaed. The same box can host your website and email too.

We talk.

A real conversation about your business, your data, and what questions keep you up at night. Up to 20 hours of consulting included — we learn your world so dokusan can too.

You try it.

30 days, free. Your data on a dedicated machine in our data center. Ask it anything. Show your team. Text it from the courthouse, the job site, the airport. See if you can stop.

You decide.

Keep it and pay monthly. Walk away and you only owe for consulting hours used at $200/hr. No contracts, no data hostage, no awkward breakup.

Solo

$2,000/mo

One brain, one person. Mac Mini with 64GB. For the lawyer, the founder, the operator who needs total recall and total privacy. Bring your own hardware or we provide it.

Team

$7,000/mo

Everyone talks to it. Mac Studio recommended. Proactive alerts, phone access for the whole team, hosted hardware, replication — everything included.

Company

$15,000/mo

Multiple locations, higher data volumes, priority onboarding, remote hands. Still cheaper than a junior analyst. Still available on Christmas.

Every tier includes hardware, software, hosting, phone access, updates, up to 20 hours of onboarding, and a 30-day free trial. Add replication for $85/mo or remote hands insurance for $50/mo. Everything else is included. No line items.

Talk to Dokusan →

Your Questions Are Discoverable. Your Data Is a Target.

The legal risk.

In February 2026, a federal judge in New York ruled for the first time that conversations with public cloud AI platforms are not privileged and are fully discoverable in court (United States v. Heppner). Every prompt you type, every answer you receive — it's all on the record, discoverable, subject to subpoena, breachable.

"Which clients are at risk?" "Who's underperforming?" "What happens if we cut this product line?" These questions reveal what you're worried about, what you're planning, and what you don't yet know. dokusan keeps all that on your private hardware. No API calls to third-party servers. No conversation logs in someone else's data center. No training data contributions. The question and the answer stay with you.

The supply chain risk.

Your current database is likely open source — PostgreSQL, MySQL, MariaDB, SQLite, CockroachDB, MongoDB, CouchDB, Redis, Valkey, RocksDB, Elasticsearch, OpenSearch, Meilisearch, InfluxDB, TimescaleDB, ClickHouse, DuckDB, Neo4j, Qdrant, Milvus, Chroma, Weaviate. If you use any of these you are trusting unpaid upstream maintainers, third-party contributors, and transitive dependencies with your data. A compromised commit, a hijacked package, a maintainer who burns out and hands off the keys — that's not hypothetical, it's your supply chain risk.

Trivy + LiteLLM (March 2026).

Trivy — the most widely used open-source vulnerability scanner — was compromised by a group tracked as TeamPCP. They rewrote 76 of 77 version tags to point to malicious commits, published a backdoored release, and planted persistent backdoors on developer machines. LiteLLM's CI/CD ran Trivy without a pinned version. The attackers got LiteLLM's PyPI publishing token. LiteLLM is downloaded ~96 million times per month. Malicious versions were live for ~5 hours. The malware harvested AWS/GCP/Azure tokens, SSH keys, Kubernetes configs, API keys, and database passwords. The attackers claim to have exfiltrated over 300GB of data.

Axios (2026).

A North Korean state actor social-engineered one unpaid maintainer and got a remote access trojan onto potentially millions of machines. Axios is downloaded ~100 million times per week. The attackers impersonated a real company, set up a fake Slack workspace, then moved to a Teams meeting with multiple apparent participants. Three separate payloads were pre-built. Both release branches were hit within 39 minutes. Every trace was designed to self-destruct.

One person, or a state actor, can social-engineer their way into a two-person project and compromise millions because an unpaid maintainer handed off a package to a stranger. You will hear about it when they encrypt your data and you have to learn how to buy bitcoin to get your company back.

Data.

Trust.

Hire dokusan.

We talk.

You try it.

You decide.

Today's Databases Make You Think Like a Database.

No database. No SQL. No admin. No middleman.

Understands.

Proactive.

Connected.

Trustworthy.

Learns.

Under the Hood.

問 Conversational

察 Proactive

速 Performance

守 Security

隠 Trust

流 Workflow-ready

Your Questions Are Discoverable. Your Data Is a Target.

The legal risk.

The supply chain risk.

Trivy + LiteLLM (March 2026).

Axios (2026).