# Research: Amicus vs. Warp.dev & Terminal AI Agents
## Overview
This report compares Amicus (The Synapse Protocol) with existing terminal-based AI solutions like Warp.dev, Claude Code, and Gemini CLI, focusing on coordination and collaboration.
## Feature Matrix
| Feature | Amicus (Synapse) | Warp.dev | Claude Code | Gemini CLI |
| :--- | :--- | :--- | :--- | :--- |
| **Primary Goal** | Multi-agent Context Bus | Collaborative Terminal | Project Automation | Workflow Integration |
| **Coordination** | State Persistence (Bus) | Shared Sessions | Model Context Protocol | GitHub PR Actions |
| **Agent Agnostic** | Yes (Gemini, Claude, etc) | No (Warp AI) | No (Claude) | No (Gemini) |
| **Persistence** | Permanent Workspace State | Session-based | Project-based | Action-based |
| **Long-Running** | Heartbeat/Monitor | Ambient Agents | CLI-based | Background Jobs |
## Key Differences
### 1. The "Context Bus" vs. "Siloed Sessions"
Warp.dev excels at **real-time session sharing** between humans. However, its AI agents are often tied to the specific terminal session. Amicus provides a **persistent filesystem-backed state** that allows a Gemini node to hand off to a Claude node seamlessly via the "Synapse Protocol."
### 2. Multi-Agent Orchestration
Warp recently introduced "Ambient Agents," but Amicus's **Phase 5** vision of a self-managing cluster with a **Bootstrap Manager** goes further into autonomous operation. Amicus is designed to manage *other* agents, regardless of their underlying model.
### 3. Integration with Existing Tools
Claude Code uses MCP to connect to enterprise data. Amicus *is* an MCP server, meaning it can be plugged into Claude, Gemini, or any other MCP-compliant client to provide a shared memory layer.
## Effectiveness for Amicus
Amicus is uniquely positioned as the **glue** between different AI providers. While Warp provides the *environment*, Amicus provides the *memory* and *coordination protocol* for a heterogeneous swarm of agents.
## Recommendation for Phase 5
- **Warp-like feature to adopt:** "Ambient" monitoring (already in progress with `monitor.py`).
- **Differentiation:** Double down on the **Bootstrap Manager** role to provide a "single point of control" for the user to interact with a complex multi-agent cluster.
