ToolExperimentalA2AMCPAG-UI

ghost

Name: ghost
Rating: 3.0 (46 reviews)
Author: ghostapp-ai

by ghostapp-ai

Local-first Agent OS with MCP/A2A protocols and on-device inference

Rust

Updated Oct 19, 2026

Stars

Forks

View on GitHub

Summary

Provide a private, local-first Agent OS that runs AI agents and indexes local files for semantic search. Implements a full protocol stack (Model Context Protocol (MCP), AG-UI, and Agent-to-Agent Protocol (A2A)) so agents, UIs, and tools interoperate on-device with zero cloud and zero telemetry. Distinctive for Rust-based desktop builds and on-device inference (gguf-compatible models) enabling offline agent workflows and tool-calling.

The Value Proposition

As agents move data and delegate tasks across other agents, keeping those interactions private and auditable matters for trust. Ghost focuses on local-first execution and open protocols so teams can evaluate agent behavior without sending sensitive context to the cloud. That makes it useful for building reproducible agent track records and observing agent failure modes in a controlled, privacy-preserving environment. This aligns with best practices like Tool Use Pattern to safely orchestrate capabilities.

Target Use Cases

Developers building privacy-sensitive, local multi-agent workflows who need open protocol compatibility and on-device model execution. For teams prioritizing openness and interoperability, consider supporting standards like the Open Agent Specification (Agent Spec).

How It's Used

Inspect and search local code/data and run agents without cloud telemetry
Prototype agent-to-agent workflows using MCP/A2A protocols on-device
Evaluate agent behavior and failure modes privately to build an agent track record
Connect desktop UIs and tool-callers via AG-UI for reproducible local testing

See related protocols

Standards this tool supports