Meet tvaras.The AI-native data platform that plugs to your stack.
Immediate answers for stakeholders. A force multiplier for your data.
Or schedule a demo with the team.
| Source | Last sync | Status |
|---|---|---|
| stripe_charges | 2h ago | ✓ |
| paypal_events | 45min ago | ✓ |
| app_sessions | 12h ago | STALE |
app_sessions is stale — last sync was 12 hours ago. Triggering incremental run now.
Chats. Playgrounds. Jobs.
Every way a data team works, in one place. All three surfaces share the same knowledge of your warehouse, your dbt models and your metrics — and the same permissions.
Chats
Plain-language questions against your warehouse, answered with the SQL, the chart and the explanation. Drill in, change the range, follow up. Turn any answer into a schedule that lands in Slack, Jira or your inbox.
brand_search_US at €67 vs €35 target. Mobile CVR 1.2% vs 3.8% desktop.- Verified against your warehouse, not guessed
- Fluent in dbt, your BI layer and your metric names
- Automations: set it once, get it where you work
Playgrounds
Describe the tool you need — a campaign monitor, a reconciliation screen, a lightweight CRM — and get a real, persistent app at /pg/your-slug. Wired to your data, shareable, permission-aware, version-controlled.
Total
€48.2K
vs prev
+12%
Channels
3
- Charts stay editable in the builder, not frozen in code
- Own datastore, file uploads, scheduled snapshots
- Templates and an element library your team grows
Jobs
A chat that owns a git branch. Start from a message or a Linear, ClickUp or Jira task; the assistant investigates, proposes a plan, edits dbt models and code on its own branch, runs your checks and hands you a diff to review, commit and merge.
- Goal, acceptance criteria and a verify command, agreed up front
- Autopilot works until the goal is met or it needs you
- Changes, Files, Git and Checks panels next to the chat
From conversation to interactive app
Most data tools stop at the chart. tvaras keeps going. Describe what you want — "a daily ad-spend dashboard with drilldown into top campaigns" — and a real, persistent app shows up at /pg/your-slug, wired to your warehouse and ad accounts. Filterable, shareable, version-controlled. Dashboards as small apps.
Build me a daily ad-spend dashboard. Stacked area by channel, drilldown into top campaigns.
Total
$48.2K
vs prev
+12%
Channels
3
A data platform your team can run entirely from a phone.
Not a read-only companion app. The full workbench: chat with the assistant, open a diff, edit or create a dbt model in an editor with a proper key bar, commit, push, merge, approve a plan or an automation — from the train, the couch or the conference hall. Installs as an app and pushes a notification when a job needs you.
- Edit and create dbt models
- Review diffs hunk by hunk
- Commit, push, merge, open pull requests
- Swipe between jobs, approve plans and automations
- customer_ltv net of refunds
- tests pass
- explore dimension visible
Chat + diff
Editor with key bar
- Mmodels/marts/dim_customer.sql
- Mmodels/marts/schema.yml
- Mmodels/marts/_lightdash.yml
lightdash: expose Lifetime value dimension on Customers
Git: commit · push · merge
Creating a job takes one message. Or one ticket.
No forms, no setup. A job is a chat that is goal-oriented from the first line — and it keeps a real branch, a real diff and your real checks next to the conversation.
- 1
Start
New Chat ▸ Job, or pick a task from Linear, ClickUp or Jira. The task moves to your team's "in progress" and gets a link back to the job.
Start a job from a taskLinearClickUpJiracustomer_ltvpaste a URL or keyDATA-412TodoAdd customer_ltv to dim_customer
DATA-409TodoRefund events missing from stg_stripe
DATA-401BacklogWeekly churn report → Slack
DATA-412·TodoJ. ReyesAdd customer_ltv to dim_customer
Finance wants lifetime value per customer in the BI explore. Sum of net revenue from
fct_orders, refunds excluded. Add a test.Recent comments“Net of refunds, please — same as the finance sheet.” — M. Kaur
Start jobOpen existing job→ moves to In progress - 2
Plan
The assistant reads the task and the repository first, then proposes a goal, acceptance criteria, a verify command and an approach. Nothing changes before you accept.
Started from Linear DATA-412job_task_getRead models/marts/dim_customer.sqlGrep net_revenuejob_plan_proposePlan proposed — your reviewGoalExpose
customer_ltv(net of refunds) ondim_customerand in the Lightdash explore.Acceptance criteria- dim_customer.customer_ltv = Σ net_revenue from fct_orders, refunds excluded
- not_null + non-negative tests pass
- Lightdash dimension customer_ltv visible in the Customers explore
Verify commanddbt build -s dim_customer+
ApproachAdd CTE in dim_customer.sql · schema.yml tests · lightdash meta · run build
- 3
Work
On its own branch it edits dbt models and code, runs the checks, ticks criteria. Autopilot keeps going until the goal is met or it needs you. Every hunk is yours to keep, revert or discuss.
DATA-412 customer_ltvIn progressDATA-412·In progress working · autopilot onjob/customer-ltv-412Goal · 2 / 3verify: dbt build -s dim_customer+- customer_ltv net of refunds
- tests pass
- explore dimension visible
Edit dim_customer.sqlEdit schema.ymlBash dbt build -s dim_customer+Build passed — 4 models, 6 tests. customer_ltv is now net of refunds. Adding the Lightdash dimension next.Round it to 2 decimals in the explore, and label it “Lifetime value”.Message the job…↵Changes3FilesGitChecks dbt build - 4
Review & ship
Commit with an AI-drafted message, push, merge or open a pull request. The linked task follows the job into review and done.
In review· snapshot frozen · goal 3 / 3 Approve & mergeOpen a pull requestRequest changesFiles · 3- models/marts/dim_customer.sql +3 −1
- models/marts/schema.yml +9
- models/marts/_lightdash.yml +4
Commits · 2- a41c9e dim_customer: add customer_ltv net of refunds
- f07b22 lightdash: expose Lifetime value dimension
Checks- dbt build -s dim_customer+ · 41s
- sqlfluff lint · 3s
- lightdash validate · 12s
Linked taskDATA-412·QA→ on approveDATA-412·Donesquash → main
Everything else a data platform needs.
One pipeline from your sources to your team — shared by all three surfaces. Click a node.
Your sources
tvaras
Surfaces
Where it lands
Knowledge
Auto-detected exposures from your BI tool and dbt docs, a metrics dictionary, and memories of your conventions and corrections. Every answer is checked against the warehouse before it reaches you.
Built for every team that touches data.
Founders ask, marketers attribute, data teams ship — on the same platform, with the same answers.
For founders
A CFO-style weekly read, on demand. Cash, runway, board metrics — without a single ping to the data team.
Read the use casesFor data teams
A BI engineer who lives in dbt, git and your BI tool. Drains the ad-hoc queue, ships models as reviewable jobs — and lets you approve them from your phone.
Read the use casesFor marketing
An attribution analyst who reads every marketing channel and your warehouse. CAC by campaign, creative tests, channel mix.
Read the use casesSimple, transparent pricing
One plan, self-hosted on your infrastructure. Try it free for 14 days.
Self-Hosted
Runs on your own infrastructure
+ API costs
- Up to 50 users
- Chats, Playgrounds and Jobs
- Your own infrastructure
- Full data sovereignty
- Multi-warehouse support
- dbt + BI integration
- Slack, Jira, Linear, ClickUp
- Scheduled automations
No credit card required
Enterprise
For scaling organizations
Tailored to scale + integrations
- Everything in Self-Hosted
- Unlimited users
- Custom connectors
- Your choice of LLM
- Unlimited marketing-channel accounts
- Granular permissions & roles
- Priority support & onboarding
The free trial runs self-hosted on your own infrastructure and currently requires Google BigQuery + dbt. Other warehouses and transformation tools are coming soon.
Frequently asked questions
Straight answers, no fluff.
It complements them. A general assistant starts blank, or knows only what one person told it. tvaras holds your whole company’s knowledge — and who’s allowed to see what — and connects to all your tools. It already knows your dbt models, your metrics, your warehouse, so "CAC by channel this month?" returns a real query, a chart, and the explanation. A proprietary engine models and combines your data; a verification engine checks every answer against your warehouse, so you get facts, not hallucinations. Power users can also reach it over MCP.
No. Skills are one layer. Add purpose-built tools (warehouse queries, dbt runs, BI lookups, Slack and Jira actions, scheduled jobs), a knowledge layer that learns your metrics and conventions from real queries and corrections, and three surfaces to act on them — Chats for answers, Playgrounds for apps, Jobs for shipping changes to your repository. It’s a system on your data, not a folder of prompts.
You do. tvaras runs self-hosted in your own infrastructure — your data never leaves it, nothing for the sub-processor list. Read-only by default, every query logged for audit, granular permissions, no model trained on your data. Bring your own LLM, so you choose which provider ever sees a prompt.
No. It gives them their time back. tvaras clears the ad-hoc request queue so stakeholders get answers instantly. Your team stops being the SQL help desk and ships the hard work. An extra pair of hands, not a replacement.
Any warehouse. Fluent in dbt and your BI layer (Lightdash, Looker, Metabase). Use something we don’t support yet? We build the connector fast for your use case. A short turnaround, not a roadmap item.
Fast. No new warehouse, no migration — it plugs into the stack you already run. Connect your warehouse and dbt project, ask questions, get charted answers the same day. It sharpens as it learns your metrics. First useful answers in hours, not weeks.
Yes — and not just read dashboards. Jobs give every change its own branch, so from a phone you can ask for a new dbt model, edit the SQL in an editor with a proper key bar, run the checks, review the diff hunk by hunk, commit, push and merge or open a pull request. Playgrounds are responsive, automations are approved from the chat, and push notifications tell you when a job needs you. It installs to the home screen as an app; nothing routes through us.