From the source
- Taxbit built Taxbit Assist, a cited, permission-aware AI assistant embedded in its digital-asset tax compliance dashboard, on
Amazon Bedrock AgentCore.
- The knowledge needed to use the dashboard was scattered across the public website, API documentation and help-center articles, and a meaningful share was never written down, so users either dug for answers or filed support tickets.
- Four constraints shaped the build: every answer must trace to a source the user can open, the assistant must respect dashboard roles and organization entitlements, it must explain how Taxbit computes cost basis without giving tax advice, and it must support regional data residency from a single codebase.
- The agent is a small Python application with a system prompt, a model and eight narrow tools, including knowledge base search, two access checks and four support ticket tools; it has no read access to customer tax data.
- Identity is never handled by the model: email, organization ID and roles come from the trusted calling service and are bound into each tool through a closure, so a prompt injection cannot imitate the trusted context line.
- Taxbit says more than 55 percent of users who ask Assist a question return to ask another within 24 hours, according to its internal production analytics.






