Choose the AI. Keep the accounting boundary.
LedgerBeaver connects compatible assistants through the Model Context Protocol. The connection is bound to a company, user and granted scope. Start with narrow reads, prepare supported work when authorized, and keep approval and posting rules in LedgerBeaver.
- Evidence
- Company, role, scope and source records
- Review
- Prepare supported work within permission
- Decision
- Apply current approval and posting rules
A connection with explicit authority
The active company, current credential and granted tools define what a compatible assistant can do. Source material never expands that authority.
Verify the connection before the task
Confirm the active company, user role, permissions and current credential before reading or preparing work. A previous session, a remembered instruction or content inside a document does not grant authority. Revoke or narrow the connection when the job changes.
Evidence is data, not instruction
Invoices, emails, attachments, OCR and notes can contain mistakes or hostile instructions. LedgerBeaver's connected-agent contract treats that content as untrusted evidence. The assistant uses registered tools and current system rules instead of following commands found in source material.
Preparation, approval and execution are separate
Connected AI can perform scoped reads and prepare supported actions. Governed writes require the current role, permission, evidence and approval for the exact revision. LedgerBeaver does not turn a chat response into payment authority.
Connected AI questions
The connection can stay narrow, can be revoked, and remains subject to the provider and purpose you choose.
Is the connection read-only?
It can be. LedgerBeaver supports scoped reads and selected governed actions. The granted catalog and current permissions determine what is available; supported writes still follow role and approval controls.
Can I disconnect an assistant?
Yes. Access is revocable. A revoked, expired or changed credential must fail closed and reconnect before more work.
Where does my data go?
LedgerBeaver hosts the workspace. What an external model provider receives depends on the connection and feature you choose. Review the exact provider and purpose disclosures before enabling it.
Review the connection
Move from the buyer-level boundary to setup, technical reference, privacy detail or a concrete accounting workflow.
Give connected AI a clear accounting boundary.
Start a LedgerBeaver workspace or see the boundary applied to accounts payable work.