Governed data, right where Claude works
SoTag is the governed-data teammate in your Slack. Claude — Anthropic's Claude Tag— reasons, writes and builds in the same channel, but by design it can't see your business data. SoTag can: governed, audited, never invented. It hands those numbers into the thread so Claude's judgment is grounded in figures you can trust.
Two agents, one channel, one conversation — and a human on every action that moves money.
Early access — joining a pilot Slack workspace, not a self-serve install.
Bill @Claude @SoTag — should we run a 30% weekend promo to lift repeat purchase?
Claude APP
30% is steep — protect margin, and don't train people to wait for sales. I'd target lapsed members only. The numbers are @SoTag's call →
SoTag APP
from the semantic layer · audited · illustrative
Reasoning is only as good as the data under it
Claude Tag is a remarkable teammate — it thinks through a decision, writes the copy, builds the dashboard, ships the code. But it is blind to your business by design: it sees the Slack thread, not your systems. So in the channel where decisions get made, the smartest agent in the room is missing the one thing that makes the answer real — your numbers.
SoTag is that missing half.It's the one agent wired into your governed data — certified metrics, member operations, human-approved sends — and it puts those figures on the table right where Claude is already working. You get Claude's judgment grounded in trustworthy data, without handing a general-purpose agent the keys to your systems.
Two agents, two jobs, one thread
Think of it as an outside brain and an inside hand. Neither can do the other's half — together they carry a question from “what's happening?” all the way to a built, measured result.
Claude
Anthropic's Claude TagReasons through a decision, models the economics, drafts the copy, builds an HTML dashboard, writes the code, opens the PR. Brilliant at turning inputs into work — but it only knows what's in the thread. Your business data is invisible to it by design.
- Strategy & judgment
- Writes code · opens PRs
- Builds dashboards & docs
SoTag
the governed-data teammatePulls governed metrics, watches them every day, runs member operations, and — behind human approval — sends the coupon. Everything on the same certified semantic layer as SocialHub and Flash, with a full audit trail. It's the one agent wired into your real numbers.
- Governed, audited metrics
- Proactive daily insights
- Member ops · human-approved sends
One reasons and builds; the other knows the numbers and can act on them. Put them in the same channel and the gap between insight and execution closes.
No integration to wire up — it happens in the thread
There's no special pipe between the two agents. They simply share the conversation — and SoTag is built to read it, contribute its half, and pass the baton.
Reads the moment
SoTag sees the recent thread — including what Claude just said — and answers where the conversation actually is, building on it instead of restarting from scratch.
Does its half, then hands off
When the ask is outside its lane — an HTML dashboard, a code change, a PR — SoTag first pulls the governed numbers, then @mentions Claude in the thread and passes those exact figures along.
Claude builds on real data
Claude turns SoTag's certified numbers into the artifact or the decision. The figure becomes a shared, trusted input the whole team works from — not a number trapped in one more tool.
Reading top-to-bottom, it feels like one conversation — a teammate asks, Claude reasons, SoTag grounds it in real numbers, Claude builds on them. Not a row of bots answering isolated prompts.
Data-grounded AI — without widening who can read your data
The natural worry with more agents in the channel is more doors into your data. The opposite is true here: keeping Claude data-blind and routing every number through SoTag is the cleaner, safer design.
Claude never touches your data
Claude sees only the figures SoTag chooses to surface in the thread — never your database, your credentials, or your systems. Adding a brilliant general agent to the channel does not add a second door into your data.
One governed path, one audit trail
Every number flows through SoTag's tenant-scoped, certified semantic layer with a per-call audit trail. There is one place data comes from, and one place it's recorded — no matter how many agents are in the room.
Grounded, never invented
SoTag's figures come from the certified semantic layer. If there's no data for what was asked, it says so — so Claude reasons on real numbers, not a helpful guess.
Tenant isolation · least privilege · full auditability — extended to the whole channel, not just one tool.
A colleague that watches, answers, and acts
SoTag does more than answer when asked. It watches your headline metrics, speaks up when something moves, and — with your approval — can act on what it finds. Each capability below is real today in pilot.
Ask in plain language, get governed numbers
@mention SoTag or use /so. Answers come from the certified semantic layer — GMV, active members, redemption rate, repeat-purchase rate, orders — as multi-metric scorecards with time windows, store / channel / tier breakdowns, and period-over-period change. It never invents a figure; if there's no data, it tells you.
★ Proactive insights — the north star
A daily scan watches your headline metrics and proactively flags a meaningful move — say, redemptions down more than 20% versus the prior period — right in the channel, where Claude and the team can react to it together. It stays silent when nothing's wrong: dedup plus a per-channel daily cap mean it never spams. Run /so scan to check on demand.
Weekly business digest
Every Monday SoTag posts a full scorecard with week-over-week deltas, so the whole channel — people and agents — starts the week on the same numbers. Need it sooner? /so report, anytime.
Channel memory
Tell it once — /so remember “always exclude the warehouse store” — and SoTag honors that standing instruction in every future answer for the channel.
★ From insight to action — with a human in the loop
When a problem surfaces, SoTag proposes a concrete, adopt-able play — for example, a win-back coupon to the at-risk segment — or you can ask it directly in the channel to send a coupon to a member or segment. A confirmation card shows exactly who it reaches and the economics; a person confirms; SoTag issues the coupons through the governed write path — idempotent, compliance- and budget-checked, fully audited — and follows up with the redemption and GMV effect. Detect → recommend → approve → send → measure: the loop, closed.
The loop, closed — with a human in it
The point isn't a faster dashboard — it's that the team can carry an insight all the way to a measured result, pausing for a person at the one step that moves money.
The daily scan flags a meaningful move in a headline metric.
SoTag proposes a concrete, adopt-able play in the thread.
A confirmation card shows who it reaches and the economics; a person confirms.
It issues through the governed write path — idempotent, compliance- and budget-checked.
Days later it returns with the redemption and GMV effect.
SoTag prepares and measures; a person approves the send. It proposes plays — it does not blast.
Just type in the channel
@mention SoTag or use /so — often in the same breath as @Claude. A few of the things teams do in pilot:
Ask anything
- @SoTag GMV last week by store, top 3
- @SoTag redemption rate this month vs last
- /so active members by tier
Hand off to Claude
- @SoTag repeat-purchase rate + dormant 60d count
- …then @Claude build a dashboard from those
- SoTag passes the governed figures along
Act & shape
- @SoTag send a $10 coupon to alice@example.com
- Adopt a proposed win-back play
- Confirm before anything is sent
- /so remember “exclude the warehouse store”
It won't make up a number.Every figure comes from SocialHub's governed semantic layer. If there's no data for what you asked, SoTag tells you — it never guesses to be helpful, and Claude never has to reason on a fabricated figure.
Governed the same way as the rest of the platform
SoTag uses the same MCP server and tenant-scoped keys as SocialHub's developer platform — autonomy you can actually govern.
Tenant-scoped keys
Every key is scoped to your team only; SoTag can never reach another tenant's data.
Per-tool authorization
Each tool is authorized individually — read and write scopes are granted, not assumed.
Write-scope guard
Money-moving writes pass a server-side guard (value cap + rate limit) and fail closed if anything is off.
Human approval + kill-switch
A person confirms every money-moving action, and a kill-switch can stop the agent at any time.
Every call is recorded in a per-call audit trail. Tenant isolation · least privilege · full auditability.
Questions, answered
What is SoTag?
SoTag is an AI marketing and member-operations colleague that lives in your Slack. It runs on the same governed MCP and certified semantic layer as SocialHub and Flash, so it can pull a trustworthy metric or run an approved operation right in the channel — and, crucially, it's the one agent in the room that can see your real business data.
How does SoTag work with Claude (Claude Tag)?
They share the same Slack thread. Claude — Anthropic's AI teammate — is excellent at reasoning, writing and building, but by design it can only see what's in the thread, not your business systems. SoTag reads the thread too, so it builds on what Claude just said, and when a request needs the numbers it pulls them from the governed semantic layer and hands them to Claude in the same thread. Claude's judgment, grounded in SoTag's certified data — one coherent conversation.
Does Claude see my business data?
No. Claude only sees the specific figures SoTag chooses to surface in the thread — never your database, credentials, or systems. That's the point: you get data-grounded AI work from a general agent without giving it the keys to your data. All numbers still flow through one governed path, with one audit trail.
Is SoTag available today?
SoTag is in early access — it is real and running in pilot Slack workspaces today, not a concept. Money-moving actions run behind human approval and are rolling out gradually. To join a pilot, book a demo or talk to us.
Can SoTag send coupons on its own?
No. SoTag can detect a problem and propose a play, but a person must confirm before anything is sent. The confirmation card shows exactly who the action reaches and its economics, and the send runs through a governed, audited write path with a value cap and rate limit.
Can SoTag help with customer complaints?
Yes — it's a natural fit for service recovery. When a complaint surfaces in Slack, a team member can ask SoTag to send a make-good coupon to that customer right in the thread — for example, "send a $10 apology coupon to alice@example.com". SoTag resolves the member, shows a confirmation card with the audience and economics, and on a person's approval the coupon lands in the customer's wallet through the governed, audited write path.
How is SoTag governed?
SoTag uses the same MCP server and tenant-scoped keys as SocialHub's developer platform: per-tool authorization, a write-scope guard that fails closed, a kill-switch, human approval on every money-moving action, and a per-call audit trail. Metric numbers come from the certified semantic layer — the agent never invents figures.