Skip to content
View ellyseum's full-sized avatar
  • 03:28 (UTC -05:00)

Block or report ellyseum

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don't include any personal information such as legal names or email addresses. Markdown supported. This note will be visible to only you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
ellyseum/README.md

Ellyseum

Agent infrastructure for the real world.

Coordination and security layers that multi-agent systems need to work in production. Not demos. Not wrappers. Infrastructure.


The Problem

When you give multiple AI agents the same task, they run blind. No awareness of each other. They overwrite files, duplicate work, pick conflicting names, and race for the same resources. The standard fix is an orchestrator — a central brain that assigns tasks and sequences work. Orchestrators are bottlenecks, single points of failure, and they have to understand the problem well enough to decompose it correctly.

Coordination should be bottom-up. Agents should self-organize through shared primitives, not wait for instructions from a planner that might be wrong.

Projects

MCP server + REST API for multi-agent coordination. Claims, shared state, events, and discoveries — the primitives agents need to self-organize without an orchestrator. PIA Guard scans every write, read, and snapshot for prompt injection. On by default.

v2.0: pluggable backends (memory, SQLite, Redis), namespace isolation, async interfaces. Scales from a single process to distributed Redis-backed clusters. Next: WebSocket push and the control plane for running agent fleets at scale.

Deterministic prompt injection scanner. Pattern-based, no LLM required, sub-millisecond. Defends against coordination injection attacks — where a compromised agent poisons shared state that other agents trust. Tested against 18 models with an average 49% vulnerability rate. Zero dependencies. On npm.

Visual context for Claude Code — clipboard reading, screenshots, and video analysis via Docker. Lets coding agents see what you see.

Secrets handling for Claude Code sessions. Self-destructing scripts that source credentials, run commands, and delete themselves. Secrets never appear in session logs.


Direction

The long-term vision is an Agent Coordination Protocol — a framework-agnostic spec for agent-to-agent coordination, like MCP but for the space between agents instead of between agents and tools.

Namespaces secured: @agentcoordinationprotocol on npm, agentcoordinationprotocol on GitHub.


About

Built by Jocelyn Ellyse. 20+ years shipping infrastructure that can't go down — BGP Anycast scrubbing networks, real-time DDoS mitigation at Akamai, HIPAA-compliant healthcare platforms, high-throughput data pipelines. Now applying that reliability obsession to AI systems.

Email LinkedIn Blog

Popular repositories Loading

  1. ellymud ellymud Public

    TypeScript MUD server with Telnet/WebSocket connections, React admin panel, MCP API for AI agents, and flexible storage (JSON/SQLite/PostgreSQL). Event-driven combat, state machine architecture, an…

    TypeScript 4 1

  2. visual-storytelling visual-storytelling Public

    🎨 AI-powered visual storytelling platform using Google Gemini for character creation, scene composition, and image generation. Built with React, Express, SQLite, and deployable to GCP Cloud Run.

    TypeScript

  3. InfiniteTalk InfiniteTalk Public

    Forked from MeiGen-AI/InfiniteTalk

    ​​Unlimited-length talking video generation​​ that supports image-to-video and video-to-video generation

    Python

  4. avatar-animate-from-video-mvp avatar-animate-from-video-mvp Public

    Docker-based headless Blender microservice with CUDA GPU support for automated rigging and animation retargeting

    Python

  5. frankmocap frankmocap Public

    Forked from facebookresearch/frankmocap

    A Strong and Easy-to-use Single View 3D Hand+Body Pose Estimator

    Python

  6. hand_object_detector hand_object_detector Public

    Forked from ddshan/hand_object_detector

    Project and dataset webpage:

    Python