Tacit

•
152
•

Safeguards AI agents by enforcing capability-based access control and preventing information leakage through a type-checked Scala environment.

TACIT (Tracked Agent Capabilities In Types) is a sophisticated safety harness designed for AI agents, leveraging Scala 3's advanced capture checking type system. It enables agents to execute code in a secure, sandboxed environment where access rights, effects, and information flow are statically tracked and enforced. By providing an MCP (Model Context Protocol) interface, TACIT integrates seamlessly with various agents, ensuring that operations like file system access, process execution, and network requests adhere to explicitly granted capabilities, thus preventing forging access rights, exceeding budgets, or leaking sensitive data.

Características Principales

01Statically tracks and enforces capability safety for AI agents using Scala 3 capture checking.
02Provides a secure, sandboxed Scala REPL for both stateless and stateful agent code execution.
03Offers a capability safety library as a typed API for controlled interaction with file system, processes, and network.
04Ensures information flow control and prevents data leakage using the `Classified[T]` type system.
05Integrates with any MCP-compatible agent via a standard JSON-RPC server for seamless deployment.
0615 GitHub stars

Casos de Uso

01Safeguarding AI agents against unauthorized resource access and malicious actions.
02Processing sensitive or classified data within AI agents without risk of information leakage.
03Enforcing strict operational boundaries for AI agents interacting with real-world systems.