Connectors
Give your agents tools that reach outside the workspace — into Asana, PostHog, the open web, and their own files.
Connectors are the tools your agents pick up to do real work: reading and writing in an outside system, pulling a page off the web, or keeping notes as they go. When a connector supports it, an agent can act — not just look.
What agents can connect to today
- Asana — agents can read your workspaces, projects, and tasks, and act on them: create a task, update or complete one, and add comments. This runs on your own Asana access token, so agents see and change exactly what that token allows.
- PostHog — agents can read your product analytics: which events fire and how often, trends over time, conversion funnels, cohort sizes, your saved insights, and HogQL queries. Read-only, and aggregates only — see PostHog below.
- The open web — agents can fetch any URL, sending a GET to read a page or a POST to hand data to an endpoint. Useful for pulling reference material or calling a simple API mid-task.
- Their own files — every agent has a
/workspaceit can read from and write to: create files, edit them in place, and organize them into folders. This is where drafts, exports, and generated documents live during a job. - Stripe — agents can read your Stripe account: customers, subscriptions, invoices, charges, payment links, and an MRR, new and churn summary. Read-only, with emails and card details masked — see Stripe below.
For agent notes and long-term recall across jobs, see Memory — it's a separate, always-on capability, not something you connect.
Read vs. act
Not every connector writes back. The web and file tools read and write freely. Asana both reads and acts — creating and updating tasks and posting comments — but only within the permissions of the token you connect.
Stripe
Connect Stripe and your Queen Bee, Sales Analyst, Data Analyst and Product Manager can answer questions like "what's our MRR?", "who churned last month?" or "which invoices are overdue?" straight from your billing data.
Create a restricted key. In the Stripe Dashboard go to Developers → API keys → Create restricted key. Give it Read access on Customers, Subscriptions, Invoices, Charges, Payment Links, Prices, Products and Balance, and None on everything else. Use a test-mode key (rk_test_…) to try it out first. Secret keys (sk_…) and publishable keys (pk_…) are refused: a restricted key can be limited to read-only, a secret key can't.
Connect it. Go to Settings → Integrations → + Add → Stripe restricted key and choose Connect for: Personal (any member — your agents use your key) or Organization (owners and admins — every member's agents can use it). If an agent needs Stripe and it isn't connected, it posts a secure Connect card in the chat instead of asking for the key; a key connected from that card is saved as yours. Never paste a key into a message.
We check the key with Stripe (it must be able to read your balance) before saving it, and store it encrypted. It is never shown again or sent into an agent's workspace; Settings shows only its mode and last four characters.
Which connection is used. Your own Stripe connection first, then the organization's. Never another member's.
What agents can do (all read-only — nothing is ever written to Stripe; no refunds, charges, cancellations or new payment links):
| Tool | What it does |
|---|---|
stripe_list_customers | Customers, newest first, with a filter by exact email or created date |
stripe_get_customer | One customer and a summary of their subscriptions |
stripe_list_subscriptions | Subscriptions by status, customer, price or created date |
stripe_list_invoices | Invoices by status (e.g. open), customer, subscription or created date |
stripe_list_charges | Payments with amount, status, refunds and failure reason |
stripe_list_payment_links | Your payment links and what each sells |
stripe_revenue_summary | MRR at the start and end of a date range, new and churned subscriptions, per currency |
Lists return 25 results a page by default (100 at most) and a cursor for the next page. The revenue summary reads up to 2,000 subscriptions and says when it stopped short.
How MRR is counted. Each subscription item's price × quantity, converted to a monthly amount (a yearly price ÷ 12, weekly × 52/12, and so on). It is before discounts, coupons and tax; usage-based (metered) and tiered prices are left out and counted separately; subscriptions in a free trial, or paused after one, don't count yet. It uses each subscription's current items. New means the subscription started in the range, churned means it ended in the range. Each currency is reported on its own, never converted. Because the Stripe Dashboard applies discounts, its MRR can be a little lower than this figure.
Customer details stay private. Emails are shown as j***@acme.com and phone numbers as their last four digits; addresses are cut to the country; cards show only the brand and last four. Payment secrets, payable invoice links and receipt links are never returned. An agent can see one customer's full email only by asking for it on that customer (for example, to draft a payment reminder), and that is logged. Card details are never shown.
Cost. Stripe reads use no extra credits beyond the agent's normal work. This is your own Stripe account; it has nothing to do with how Busy Bee bills you.
Turning them on
Asana needs a credential before agents can use it: add your access token in Integrations, scoped to your organization or a single project. Connecting outside tools like Asana requires a paid plan — it isn't available on Free. The web and file tools need no setup — files are switched on for every agent, and the web tool for the roles that use it.
PostHog
Connect PostHog and your analyst and product agents can answer questions like "where do new signups drop off?" or "which features grew this month?" straight from your product data.
Connect it. Go to Settings → Integrations → + Add → PostHog API key and choose Connect for: Personal (any member — your agents use your PostHog access) or Organization (owners and admins — every member's agents can use it). Then enter:
- a PostHog personal API key (PostHog → Settings → Personal API keys; read access to query, insight, cohort, event definition and project is enough),
- your project ID — the number in your PostHog URL (
/project/<id>), - the host:
us(US Cloud, the default),eu(EU Cloud), or your self-hostedhttps://address.
We check the key with PostHog before saving it and store it encrypted; it is never shown again or sent to an agent. If an agent needs PostHog and it isn't connected, it posts a secure Connect card in the chat instead of asking for the key — never paste a key into a message.
Which connection is used. Your own PostHog connection first, then the organization's. Never another member's.
What agents can do (all read-only — nothing is ever written to PostHog):
| Tool | What it does |
|---|---|
posthog_query | Runs a HogQL SELECT, capped at 1,000 rows (100 by default) |
posthog_list_events | Lists recent events with counts and when each was last seen |
posthog_event_definitions | Lists your event definitions: descriptions, tags, 30-day volume |
posthog_funnel | Conversion through ordered steps for a date range, with an optional breakdown |
posthog_trends | Events over time, as totals or unique users, with an optional breakdown |
posthog_cohorts | Cohorts and their sizes |
posthog_list_insights / posthog_get_insight | Your saved insights and their latest results |
The Data Analyst, Data Scientist, Sales Analyst, Product Manager and Queen Bee roles get these tools automatically; give them to any other role on its Tools tab.
People stay private. Agents get numbers, not people: queries that read person tables or identifying fields (email, name, IP, distinct ID, person properties) are refused unless they only count them, breakdowns by those fields are refused, links to the people behind a number are removed, and any email address that still appears in a result is hidden. Saved insights that list individual people or raw events show their definition but not their rows.
Cost. PostHog calls are read-only, so they use no extra credits beyond the agent's normal work.
Where it fits
- Connect and manage tools in Integrations
- See connectors in action inside Web research and Documents and reports
- Give agents lasting recall with Memory