AI-powered business leaders

Build Your AI Workforce

Etherion is an open-source platform for deploying autonomous AI agent teams that work together to achieve complex business goals.

Proud member ofNVIDIA Inception Program
MIT Licensed · Source Available

The Foundation Everything Is Built On.

Etherion is open source — not as a feature, but as a guarantee. The architecture is inspectable, forkable, and deployable by anyone. It is the substrate that builders stand on. Everything shipped on top of it inherits that foundation.

Fully Inspectable

Every line — the orchestrator, the RLS policies, the MCP tool chain, the NixOS modules — is on GitHub. No black boxes. No trust-me clauses. Read it, audit it, understand it completely.

Permanently Ownable

Fork it today, run it in five years without asking permission. MIT license means no subscription cliff, no vendor flip, no surprise pricing. You own your deployment, your data, and your infrastructure. Forever.

A Platform to Build On

The architecture is designed to be extended. Add MCP tools, integrate new identity providers, plug in your own embedding models. The substrate is stable. What you build on top is limitless.

Start with the source.

Clone it, run it locally, read every line. No account needed.

View on GitHub
Enterprise Edition · Being Built Now

The Dinner Is Coming.

The open-source platform shows what is possible. The Enterprise Edition is what gets built for teams that refuse to compromise. It is being constructed now — on top of the same substrate you can already inspect.

Agent Swarm Management

Orchestrate hundreds of concurrent agent teams. Priority queues, resource governors, and cross-team coordination at scale.

Large Datapool Management

Compartmentalized knowledge bases with policy-driven access. Tenant boundaries that enterprise compliance teams can actually audit.

Restriction Policies & Access Design

Fine-grained tool allowlists, data access tiers, and role-based execution gates. Define exactly what each agent team can and cannot touch.

Evaluation UIs & Feedback

Structured evaluation workflows with human-in-the-loop feedback. Track agent quality over time. Improve through measurable iteration.

Failure Recovery Protocols

Automatic retry logic, degraded-mode execution, and root-cause tracing. When something fails at 2am, the system recovers without you.

Security, Audit Trails & Isolation

Immutable execution logs, SOC 2-ready audit trails, and hard multi-tenant boundaries. Built for organizations where data leakage is not an option.

Built for teams that can't afford to guess.

The Enterprise Edition is not a SaaS repackaging. It is a dedicated platform engineered for teams that put autonomous agents at the core of their operations. We are building it now, and we are building it right.

Register Enterprise InterestNo commitment. Early access notification only.

How It Works

From goal to execution in three simple steps

1

Delegate Your Goal

Start by describing your business objective in plain English. Go beyond simple tasks—assign complex, multi-step goals and provide the Orchestrator with the strategic context it needs.

Command Bar
Analyze our top three competitors' marketing from this quarter and generate a summary report
2

Observe the Workflow in Real-Time

Your private Orchestrator analyzes the goal, creates a plan, and assembles a team of specialist agents. Watch the entire reasoning process unfold in our transparent, Grok-style execution trace.

Interaction View
Thought:I need to identify the top three competitors...
Action:unified_research_tool
Observation:Found 3 competitors: CompanyA, CompanyB, CompanyC
Cost:$0.015
3

Receive the Result

The Orchestrator synthesizes the work of its specialist team into a single, cohesive final output that directly achieves your goal. Provide feedback to make your workforce even smarter over time.

Assistant

Competitor Marketing Analysis - Q4 2024

Based on my analysis of your top three competitors...

Key Findings:
  • • CompanyA increased social media spend by 40%
  • • CompanyB launched new content strategy
  • • CompanyC focused on influencer partnerships

Platform Design Goals

Database-Enforced Isolation

Row-Level Security policies in PostgreSQL enforce tenant isolation. Every connection sets app.tenant_id. Application code cannot bypass tenant boundaries.

Checklist-Based Orchestration

Explicit task checklists replace rigid loops. Tool requests require justification (what/how/why). Sequential and parallel execution modes. Real-time progress via GraphQL subscriptions.

Knowledge Base

Vector search runs directly in PostgreSQL using pgvector with HNSW indexes. Configurable embedding models generate high-dimensional vectors for text and images. Per-tenant tables with Row-Level Security. MinIO object storage for multimodal retrieval.

Technical Architecture

Bare-metal architecture with full infrastructure ownership. FastAPI + GraphQL API, Celery workers for async execution, PostgreSQL with pgvector and Row-Level Security, Redis for pub/sub, MinIO for object storage, and HashiCorp Vault for secrets. Declarative infrastructure with NixOS + Ansible.

Deployment Options

Architectural blueprints with drafting tools

Self-Hosted

Deploy on your own bare-metal servers or VMs. NixOS modules provision PostgreSQL+pgvector, Redis, MinIO, HashiCorp Vault, and HAProxy. Ansible playbooks for fleet management. Complete control over data residency and infrastructure with zero cloud vendor lock-in.

Local Development

Run locally with Docker Compose or `etherion up`. PostgreSQL, Redis, and MinIO containers with named volumes. Python 3.11+ virtual environment. One-command bootstrap with `etherion bootstrap`.

Production Ready

Systemd service management. Alembic database migrations. HashiCorp Vault for secrets injection. HAProxy + Nginx for load balancing and DDoS protection. Patroni for PostgreSQL HA. FRRouting for BGP failover.

Local Development Workflow

Run locally with `etherion up` or Docker Compose. PostgreSQL, Redis, and MinIO containers with named volumes. Python 3.11+ virtual environment. One-command bootstrap and teardown with Etherion CLI.

Database

PostgreSQL 16 + pgvector extension
Named volumes for data persistence
Alembic migrations for schema

Local Services

Redis container for pub/sub
FastAPI dev server with hot reload
Celery worker for async tasks

Infrastructure

MinIO for object storage
pgvector for knowledge base
HashiCorp Vault for secrets

Database-Driven Custom Agents

Agent definitions are stored in PostgreSQL with tenant isolation. Each agent has a system prompt, tool allowlist, model configuration, and execution limits. The Platform Orchestrator creates agent team blueprints through conversational interaction in the Agents Foundry UI, which is strictly separated from task execution in the Threads Dashboard.

Agent Configuration

System Prompt: Defines agent behavior and expertise
Tool Names: Explicit allowlist of approved tools
Model: LLM provider and model selection
Limits: Max iterations, timeout, temperature

Agent Teams

Composition: Multiple specialist agents per team
Pre-Approved Tools: Team-level tool allowlist
Execution Policy: Max concurrent executions and timeouts
Versioning: Track changes with parent version links

Model Context Protocol Integration

MCP tools connect agents to external systems through OAuth. All write operations require explicit confirmation. Rate limiting and circuit breakers prevent API abuse. Tool definitions are database-driven with tenant-scoped allowlists.

Technical Architecture

Dig Into the Architecture

Complete system diagrams — orchestrator flow, knowledge base ingestion, async job execution, multi-tenancy enforcement, MCP tool chain, replay system, and more.

View Architecture Diagrams

Roadmap: The Path to Extraordinary.

The foundation is solid. Two tracks ahead: open source stability and the enterprise edition that changes what AI teams can do.

Core Architecture & MVP

Complete

The foundational infrastructure, the 2N+1 Orchestrator, and the core agentic framework are built and validated.

Production Testing & Service Wiring

Complete

All services wired end-to-end (orchestrator, async workers, MCP toolchain, knowledge base, repository). Production-like environments validated, security hardened, observability instrumented.

Local Execution & Public Packages

Released

Etherion ships as fully self-contained Python packages on PyPI (etherion and etherion-tui). Install and run the entire platform locally — no cloud account required. The terminal UI provides one-command setup, OAuth provider management, agent orchestration, and live process monitoring, all backed by your own PostgreSQL, MinIO, and Redis stack.

Open Source v1.0 — Stable Release

In Progress

Hardening the open-source platform to v1.0 stability. Comprehensive test coverage, polished developer experience, and public documentation. The substrate that everything downstream depends on must be unimpeachable.

Enterprise Edition — Early Access

Upcoming

Agent swarm management, compartmentalized datapools, evaluation UIs, failure recovery protocols, and enterprise-grade audit infrastructure. Early access for teams that want to shape the product.

Jonathan Nde, Founder of Etherion

Built by an Architect.

Jonathan Nde, Founder of Etherion.

Etherion was born from a single, powerful insight: the future of AI is not just about building better tools, but about fundamentally changing our relationship with work. As a self-taught, curious, and fast-thinking systems architect, My vision is to empower users to move beyond tedious implementation and focus on what humans do best: high-level strategy and creativity. Etherion is the culmination of this vision A Platform designed not just for coders, but for architects.