HakimIno/seahorse

šŸ¦„ Seahorse: A high-performance AI Agent built with Rust & Python. šŸ¦€šŸ

Python

2

85 commits

updated Apr 5, 2026

See the code

README

Seahorse Logo

Seahorse Agent

The High-Performance, Real-Time Multi-Agent Orchestration Framework

Rust Python License: MIT

Overview

Seahorse is a next-generation AI agent framework engineered for enterprise-grade performance, safety, and scalability. By bridging the raw speed of Rust with the rich intelligence of Python, Seahorse enables true parallel collaboration among agents in a real-time, event-driven architecture.

Unlike traditional hierarchical agents, Seahorse utilizes a high-performance Pub/Sub message bus to facilitate asynchronous swarm collaboration, eliminating blocking bottlenecks and maximizing throughput.


Key Pillars

Real-Time Swarm Orchestration

Move beyond synchronous delegation. Seahorse agents communicate over a Rust-powered event-driven bus, allowing scouts, commanders, and workers to collaborate in parallel.

  • Sub-ms Latency: Native message routing with zero-copy communication.
  • Event-Driven: Reactive architecture that responds to environment changes in real-time.

Hybrid RAG & Long-Term Memory

Experience intelligence that never forgets. Seahorse integrates a dual-memory system for superior retrieval accuracy.

  • Vector Search (HNSW): Blazing fast similarity search powered by Rust.
  • Knowledge Graph: Capture complex relationships for deep contextual reasoning.

Secure Tool Sandboxing

Deploy with confidence. Seahorse executes untrusted tool code within a Wasmtime-sandboxed environment, ensuring host isolation and memory safety.


Technical Architecture

Seahorse leverages a hybrid stack designed for speed and flexibility. The system distinguishes between the high-level orchestration in Python and the performance-critical core in Rust.

Seahorse Architecture

Figure 1: High-level System Architecture

Multi-modal Knowledge Grounding

The pipeline supports sophisticated content parsing and graph-based grounding for diverse data types.

RAG Pipeline

Figure 2: Multi-modal RAG Pipeline Workflow


Quick Start

  • Rust 1.75+
  • Python 3.12+ (managed via uv or Nix)
  • Nix (Optional but recommended for reproducible environments)
  • uv (Ultra-fast package manager)

Installation

If you have Nix installed, simply run:

nix develop

Or if you use direnv, run direnv allow. This will set up the correct Rust toolchain, Python 3.12, and all system dependencies (OpenSSL, etc.) automatically.

2. Manual Setup

  1. Clone & Sync:
git clone https://github.com/HakimIno/seahorse.git
cd seahorse
uv sync
  1. Build FFI Core

    uv run maturin develop -m crates/seahorse-ffi/Cargo.toml
    
  2. Run Server

    ./scripts/dev.sh
    

Enterprise Verification

Seahorse maintains a rigorous testing standard across both layers:

  • Core Performance: cargo nextest run
  • Intelligence Layer: uv run pytest python/tests/

Built for the next wave of autonomous agents.

Built With

Contributors

HakimIno/seahorse

šŸ¦„ Seahorse: A high-performance AI Agent built with Rust & Python. šŸ¦€šŸ

Python

2

85 commits

updated Apr 5, 2026

See the code

README

Seahorse Logo

Seahorse Agent

The High-Performance, Real-Time Multi-Agent Orchestration Framework

Rust Python License: MIT

Overview

Seahorse is a next-generation AI agent framework engineered for enterprise-grade performance, safety, and scalability. By bridging the raw speed of Rust with the rich intelligence of Python, Seahorse enables true parallel collaboration among agents in a real-time, event-driven architecture.

Unlike traditional hierarchical agents, Seahorse utilizes a high-performance Pub/Sub message bus to facilitate asynchronous swarm collaboration, eliminating blocking bottlenecks and maximizing throughput.


Key Pillars

Real-Time Swarm Orchestration

Move beyond synchronous delegation. Seahorse agents communicate over a Rust-powered event-driven bus, allowing scouts, commanders, and workers to collaborate in parallel.

  • Sub-ms Latency: Native message routing with zero-copy communication.
  • Event-Driven: Reactive architecture that responds to environment changes in real-time.

Hybrid RAG & Long-Term Memory

Experience intelligence that never forgets. Seahorse integrates a dual-memory system for superior retrieval accuracy.

  • Vector Search (HNSW): Blazing fast similarity search powered by Rust.
  • Knowledge Graph: Capture complex relationships for deep contextual reasoning.

Secure Tool Sandboxing

Deploy with confidence. Seahorse executes untrusted tool code within a Wasmtime-sandboxed environment, ensuring host isolation and memory safety.


Technical Architecture

Seahorse leverages a hybrid stack designed for speed and flexibility. The system distinguishes between the high-level orchestration in Python and the performance-critical core in Rust.

Seahorse Architecture

Figure 1: High-level System Architecture

Multi-modal Knowledge Grounding

The pipeline supports sophisticated content parsing and graph-based grounding for diverse data types.

RAG Pipeline

Figure 2: Multi-modal RAG Pipeline Workflow


Quick Start

  • Rust 1.75+
  • Python 3.12+ (managed via uv or Nix)
  • Nix (Optional but recommended for reproducible environments)
  • uv (Ultra-fast package manager)

Installation

If you have Nix installed, simply run:

nix develop

Or if you use direnv, run direnv allow. This will set up the correct Rust toolchain, Python 3.12, and all system dependencies (OpenSSL, etc.) automatically.

2. Manual Setup

  1. Clone & Sync:
git clone https://github.com/HakimIno/seahorse.git
cd seahorse
uv sync
  1. Build FFI Core

    uv run maturin develop -m crates/seahorse-ffi/Cargo.toml
    
  2. Run Server

    ./scripts/dev.sh
    

Enterprise Verification

Seahorse maintains a rigorous testing standard across both layers:

  • Core Performance: cargo nextest run
  • Intelligence Layer: uv run pytest python/tests/

Built for the next wave of autonomous agents.

Built With

Contributors

Languages

Python

77.8%

Rust

15.5%

TypeScript

3.2%

CSS

2.6%