Agents suggest. You approve. Nothing cancels itself.
A wrong move here cancels real cover. So every agent carries a limit you set, every rule is a record you can read, and the one setting we cannot yet enforce is refused outright.
A bad AI moment here does not produce a clumsy paragraph — it cancels your insurance, so permission got built before intelligence.
Live today: the controls, a screen-aware assistant, 40 tools for outside AI clients, and one automatic loop that runs on rules rather than a model.
Everything below is labelled with where it actually stands in Subscriber Bot today — shipped, in progress, or on the roadmap.
AI-native by architecture, not bolted on
Four architectural choices that make AI a first-class part of Subscriber Bot rather than a chat window on the side.
Connected, not filed away
Providers, plans, renewals and invoices are kept as one connected map, so a question that spans all of them has one answer instead of five reports.
Every agent has a limit you set
Nine kinds of agent, three settings: off, suggest-only, or autonomous. New agents start at suggest-only, and only you change that.
One path, whoever is asking
The in-app assistant, an outside AI client and your own clicks pass the same permission checks and land in the same audit trail.
We refuse rather than pretend
Full autonomy is switched off and says why: the part that would enforce your rules is not finished, so the setting is blocked, not faked.
From intent to a governed action
Walk the path a request takes. Select any stage to see what happens there and what backs it.
Stage 1 of 5 · Context
It reads the screen you are on
Ask about the subscription in front of you without pasting a name or a reference number.
The assistant works out which subscription, provider, invoice, agent or workflow you are looking at before your question is sent. On a list or a dashboard it does not guess — it uses the tab you are on. So "when does this renew?" is a complete question.
What the AI in Subscriber Bot actually does
Filter by delivery status, then open any capability for the detail and what backs it. No capability is listed as shipped without something in the product behind it.
Where the AI shows up
The places AI meets the work in Subscriber Bot — and how far each one has actually got.
In-app assistant
ShippedA floating assistant on every screen, grounded in what you are looking at and running under your own permissions.
- On every screen — opening it never navigates you away from what you were doing
- Resolves the subscription, provider, invoice, agent or workflow from the screen first
- Honest limit: it operates the product; portfolio-wide reasoning is not here yet
See it as a real scenario
Every AI capability above shows up in a concrete Subscriber Bot story. Open one to read the full walk-through.
Give an agent a job — and decide how far it may goNine kinds of agent, three settings, and a readable rule behind every action one might take.Read the story
Autonomy that cannot be switched on by accidentThe autonomous setting is refused with a plain explanation until the part that enforces your rules ships.Read the story
Ask about the subscription you are already looking atThe assistant knows which record is in front of you, so a three-word question is enough.Read the story
Triage the subscription mail that actually needs youSeven categories, a visible confidence meter and a needs-action filter over one merged stream.Read the story
Point an outside AI agent at your portfolioForty tools, the same permission checks and the same audit trail as your own screens.Read the story
One question instead of five disconnected reportsProviders, plans, invoices, renewals and agents kept as one connected map of what you pay for.Read the storyAI you can actually let near your data
An AI-native product has to be governable. Here is where Subscriber Bot stands on each control — including the parts still being built.
Your account is a hard boundary
Every change is keyed to both the record and your account, then confirmed to have touched exactly one record. A reference guessed from another account finds nothing.
Permissions checked before anything runs
The same check applies whether the request comes from your browser, the assistant or an outside AI client. There is no agent-only route carrying extra rights.
Audited, with credentials stripped out
Who did what, to which record, in which account is written down for every change. Anything like a password, token, secret or key is removed before the record is kept.
No card vault, no charge path
A payment method here is only a note of which card pays which subscription. Subscriber Bot stores no card details and never retries a payment on your behalf.
Confirmation before destructive actions
Full autonomy is already refused. Still missing: a confirmation step and an undo before irreversible actions — over MCP, cancel and remove run on the first call.
AI cost tracking and evaluation sets
Connecting a source, tracking a subscription and taking in a message count against your plan. AI usage is not attributed, and there is no test set for agent decisions.
Models
Subscriber Bot calls no model of its own — no keys, no prompt library, no stored embeddings. The assistant runs on shared platform infrastructure, and multi-provider choice with bring-your-own-key is being built once for the whole family of products.
The honest answers
Both, and every item says which. Live: a screen-aware assistant, agents with nine jobs and an explicit limit, readable rules, a hard block on full autonomy, 40 tools for outside AI clients, and one automatic missed-payment loop. Not live: any model of our own, message reading, scoring, rule checking and mail scanning.
Put Subscriber Bot’s AI to work
A wrong move here cancels real cover. So every agent carries a limit you set, every rule is a record you can read, and the one setting we cannot yet enforce is refused outright.