Tacit FAQs

Question 1

What is Tacit and its primary purpose?

Accepted Answer

Tacit (Tracked Agent Capabilities In Types) is a safety harness designed for AI agents. Its primary purpose is to safeguard agents by enforcing capability-based access control, preventing information leakage, and ensuring secure execution through a type-checked Scala 3 environment.

Question 2

How does Tacit enhance the security of AI agents?

Accepted Answer

Tacit enhances security by statically tracking and enforcing capability safety, preventing agents from forging access rights or exceeding their operational budget. It also uses a `Classified[T]` type system for information flow control, effectively preventing data leakage.

Question 3

What programming language and environment does Tacit use for agent code?

Accepted Answer

Tacit uses Scala 3 with capture checking, enabling agents to write code in a capability-safe and type-checked language subset. It provides a secure, sandboxed Scala REPL for both stateless and stateful execution of agent code.

Question 4

Can Tacit integrate with existing AI agents?

Accepted Answer

Yes, Tacit is designed for seamless integration. It provides a standard JSON-RPC server and an MCP (Model Context Protocol) interface, making it compatible with any MCP-enabled agents, including popular ones like Claude Code, OpenCode, and GitHub Copilot.

Question 5

What core security guarantees does Tacit provide?

Accepted Answer

Tacit offers three main security guarantees: capability safety (capabilities cannot be forged or forgotten), effect safety (agent operations cannot exceed their defined budget), and information flow control (prevents data leakage from pure computations).

Tacit

Tacit

主要功能

使用案例

主要功能

使用案例