The data backend for Claude Code

Stop losing what
your agent
finds.

Your Claude Code can analyze your marketing data — but it can't keep it. TableBI connects every source, keeps the numbers honest, and pins your agent's analysis into a dashboard that stays live.

See the prompt →

Paste it into your Claude Code or Codex — it installs itself and walks you through connecting your first source. Prefer the web app?

$0 inference · your own Claude Code is the brain

Claude Code ● live
thinking…
artifact tablebi
Clicks
48.2k
Impr.
1.4M
Avg pos
8.3
Clicks · daily
Top queries · clicks · pos
running shoes4,210 · 6.2
trail running shoes2,830 · 8.1
best running shoes 20261,940 · 11.4
marathon shoes1,120 · 9.7
“best running shoes 2026” — striking distance
Shared · read-only
tablebi.com/d/k7f2qa
live · refreshes itself · no login
Paste & go

Your agent installs its own hands.

Paste this into Claude Code or Codex. It installs the CLI, plants the skill, and connects your account — you just answer a prompt or two.

$0 inference · your agent is the brain

tablebi — paste into Claude Code or Codex
Set up TableBI as my Claude Code's data backend — so you can connect my marketing sources, query them, and publish live dashboards for me.

1. Install:  npm i -g @tablebi/cli
2. Run tablebi install — it adds the tablebi skill so you'll know to use these commands next time.
3. Run tablebi login — it opens the browser to authorize; the first login also creates my account. Walk me through it.
4. Run tablebi context — confirm you're connected and show my sources + metric definitions.

Then tell me what's ready — always read the numbers from the CLI (never invent them), and ask before spending money or emailing a real person.
02 · Why your agent hits a wall

Claude Code is flow, not stock.

It's brilliant in the moment — and gone when you close the tab. It can't keep a source connected, can't refresh, can't keep the analysis, can't hand it to a client. And across five tools with five definitions of "conversion," it will compute a blended number — confidently, and wrong.

01

Can't hold the sources

OAuth, rate limits, schema drift across Google Ads, Meta, GA4, Search Console. The dirty work your agent shouldn't own.

02

Gets the numbers wrong

No shared metric definitions, so blended ROAS and CPA across platforms come out confidently miscalculated — the bug you can't see.

03

Can't keep it

The analysis evaporates: no refresh, no history, no link to send. Flow, never stock — so you redo it every Monday.

03 · The spine: Pipe → Ask → Pin

Connect once. Ask in two altitudes. Pin it live.

One backend under your agent. One trusted set of numbers runs through all three — so the number you ask is the number you pin is the number that refreshes. No LLM in the loop: reproducible and free.

01Pipe

Connect + unify

Google Ads, Meta, GA4, Search Console, CSV. OAuth or file, kept synced. Normalized into one trusted set of numbers — and every raw record kept, lossless.

02Ask

Two altitudes

Your Claude Code queries unified metrics (trusted, cross-channel) and raw (single-platform, full-granularity). $0 inference — your agent is the brain.

03Pin

Publish live

dashboard publish → a hosted read-only URL. Stores the query, not a snapshot — so it refreshes itself.

It's a CLI. tablebi install plants a skill into Claude Code / Codex — your agent learns the commands and your metric definitions, then drives everything in plain English. No prompt-engineering.

04 · Trust and granularity, not either/or

Trusted numbers. Full granularity. Both.

Most tools make you pick: a clean blended number you can't drill into, or raw data your agent miscomputes. TableBI keeps both — your agent moves between them in one breath. Nothing is thrown away at ingest.

unified · cross-channel · trusted

One set of numbers you can trust.

One definition of cost, conversions, ROAS — reconciled across every platform. The number you quote the client.

$ tablebi metrics --metric roas \
--group-by channel --json
Blended ROAS · by channelall sources
Prospecting4.6×
Retargeting3.9×
Brand Search0.9×
raw · native · single-platform · lossless

The platform's own fields, intact.

Every native column — Video views, Profile visits, whatever the platform reports. Full granularity, not raw SQL: structured, single-platform, and consistent.

$ tablebi schema --platform tiktok_ads
→ native: Video views, Profile visits…
$ tablebi query --platform tiktok_ads \
--select "Video views,Profile visits" \
--group-by "Ad name"
TikTok · by ad · nativesingle source
Ad nameVideo views
Hook A154,900
Hook B126,500

Same files underneath. Unified metrics for the cross-channel story; raw when you need the platform's own metric. Your agent asks schema to see what each altitude offers, then picks — you never trade trust for power.

05 · Your agent learns your data

No prompt-engineering. It interrogates the data.

An agent only knows what's in its context. So TableBI hands it the map, then lets it read the territory — install plants the skill, and a handful of bounded commands let it learn your metrics and your data before it ever queries.

tablebi install

Plant the skill

Drops a skill into Claude Code / Codex so the agent knows TableBI exists and when to reach for it. Binary on PATH ≠ agent knows.

tablebi context

Rehydrate in one call

Metric definitions, connected sources, freshness, data overview, command cheatsheet — the live manifest the agent reads first.

schema · values · sample

Interrogate the data

What fields exist (unified + native), what values a dimension takes, a few real rows. Always bounded — safe to read into context.

06 · Why believe it

The moat is trust + persistence, not AI.

"We added AI" is a moving target — everyone's bolting it on. The hard, load-bearing layer is the boring one underneath: the cleaning and unified definitions that make a number defensible, and the persistence that keeps it alive. That's what we built first.

Instead of Their gap TableBI
Claude Code on a raw DB
just point your agent at the data
Multi-source breaks; the metrics get miscomputed; nothing persists. Raw SQL is powerful and dangerous. Trust without losing granularity (raw), plus persistence (Pin) and managed connectors.
Dashboard tools
Looker Studio · AgencyAnalytics
Hand-built, GUI-first; you decode the charts yourself. Agent-driven, two altitudes, trusted numbers + the agent's read. Entry is the agent, not a GUI.
Traditional BI Heavy, expensive, needs a data model before it does anything. Agent-native, metric definitions ready, opens to value — marketing connectors out of the box.
07 · Start now

Give your Claude Code
a data backend.

Connect your sources, ask in two altitudes, pin a live dashboard. Your key, your agent, your numbers.

$0 inferenceYour agent is the brainConnect · Ask · Pin