A copilot for the apps you already run.
Connect your store, inbox or database, then just ask. It finds the answer and does the work, and checks with you before it changes anything.
Pick your stack. The copilot comes configured.
Each template arrives with the actions, guardrails and starting guidelines for that application. Connect the account and it runs — then change anything you like.
Google Workspace
Closed source- Draft and send from Gmail with the thread as context
- Read Drive and Sheets as knowledge, with citations
- Find time and book on Calendar
M365Microsoft 365
Closed source- Outlook, Teams and SharePoint in one copilot
- Excel and OneDrive documents indexed
- Entra ID sign-in and scoped permissions
WcWooCommerce
Open source- Look up orders, shipments and refund eligibility
- Answer against your own returns policy
- Refunds behind write confirmation
WordPress
Open source- Site content indexed as knowledge
- Draft, revise and schedule posts
- Plugin and user administration, read-only by default
Odoo
Open source- CRM, inventory and invoicing as actions
- Ask across modules in one question
- Writes scoped per module
BtBotble
Closed source- Catalogue, orders and customers registered
- Store content indexed for support answers
- Storefront widget in one step
Postgres
Open source- Schema introspected, relations discovered
- Safe read queries proposed for approval
- Anything of your own, with no API to write
Something else
- Any OpenAPI or GraphQL service
- MCP servers, picked up as they are
- Custom endpoints for the rest
Three steps, and none of them are prompt engineering.
Connect
Sign in to the application, or point hivein at a database or API. It introspects what is there and proposes the callable operations. You approve them, and writes stay opt-in.
Adjust
Guidelines are written in plain language, not configuration — how it should answer, what it must never do, when to hand to a person. Version them, change them, roll them back.
Publish
Test in a live playground against the awkward cases, then publish to a web widget, WhatsApp, Slack, email or your own API. One copilot, every surface.
There is one brain and a harness around it. Not a support agent handing off to a billing agent, and not a paragraph of prose standing between a model and your production data.
Closed source and open source, in one copilot.
Templates are not silos. Connect two and the copilot answers across both, in a single reply, with each fact traced to where it came from.
Neither vendor was going to build this for you. The copilot reads the thread, finds the order, checks it against your own written policy, and drafts the reply — with the refund itself held behind a confirmation.
A template is a starting point, not a ceiling.
Guidelines in plain language
Tone, escalation manners, what it will and will not discuss, how it signs off. Readable by whoever owns the process, versioned like anything else.
Guardrails on the action
Write confirmation, identity checks and named approval attach to the operation itself — so they hold on every path, not only the ones you thought of.
refunds.create($84)Asks firstYour interfaces, in the thread
Pickers, forms, tables and charts instead of asking someone to type an address into a chat box. Values come back structured and validated.
Any model, no rewrite
The model is a dropdown. Frontier, open weights or something you host yourself — switching touches nothing else you built.
Writes wait for you. A no comes with a reason.
Replayed from a live run on botble-sidekick. The copilot holds its plan because it writes to a real system, asks again before the email goes out, and re-plans when you reject it.
- 01
The plan is held
A plan that writes to a real system stops before it starts. You see the goal, the risk and every step, then confirm or reject.
- 02
Each write asks first
A tool marked ask-first shows the exact call with its values. Confirm it, edit the values, or reject it. Nothing runs until you decide.
- 03
A rejection is an instruction
Pick a reason and add a line. The copilot re-plans with what you told it instead of guessing.
Every run is a record you can replay.
What was retrieved, which action ran with which arguments, what came back, who approved it, and what it cost. When something goes wrong you replay it rather than reconstruct it from logs.
Run it on ours, or inside your own walls.
Questions people ask first.
What does free in beta actually mean?
Hosted copilots are free while we are in beta, with no card required. Self-hosted deployments are a conversation with sales, because they involve your environment and your security review.
Do I have to write prompts?
No. Templates arrive with guidelines already written, and when you want to change behaviour you change a sentence in plain language rather than editing a system prompt. There is no agent roster to design and no handoff rules to maintain.
Can it write to my systems, or only read?
Both, but writes are opt-in per source and confirmed before they run. High-value operations can require identity verification or a named approver, and the approval is recorded in the trace.
What if my application is closed source with no API?
If there is an API, OpenAPI or GraphQL service, or a database, hivein can register it. Botble and similar closed-source platforms are supported the same way as open source ones — the licence of the application makes no difference to how it connects.
Which model does it use?
Whichever you pick. Routing happens per task and switching models leaves your tools, knowledge, guidelines and permissions untouched. Self-hosted and in-house models connect through the same interface.
How long until something is live?
A template on a connected account is minutes. The longer part is deciding which actions may run unattended, which is a conversation with whoever owns the process rather than an engineering task.
Connect one application. Get a copilot this afternoon.
No card, no agent configuration, no prompt to write. Pick a template, sign in, approve the actions it may take.