evo-nexus

Disclaimer: EvoNexus is an independent, unofficial open-source project. It is not affiliated with, endorsed by, or sponsored by Anthropic. "Claude" and "Claude Code" are trademarks of Anthropic, PBC. This project integrates with Claude Code as a third-party tool and requires users to provide their own installation and credentials.

What It Is

EvoNexus is an open source, unofficial multi-agent operating layer built around the Claude Code CLI protocol — but not locked to any single LLM provider. It runs natively on Anthropic's claude CLI by default, and can transparently switch to OpenAI, Google Gemini, OpenRouter (200+ models), AWS Bedrock, Google Vertex AI, or Codex Auth via OpenClaude. Same agents, same skills, same workflows — your choice of backend.

It turns a single CLI installation into a team of 38 specialized agents organized in two ortogonal layers — 17 business agents (operations, finance, community, marketing, HR, legal, product, data, learning retention) and 21 engineering agents (architecture, planning, code review, testing, debugging, security, design, cycle orchestration, retrospective — 19 derived from oh-my-claudecode, MIT, by Yeachan Heo + 2 native: Helm and Mirror). The engineering layer follows a canonical 6-phase workflow documented in .claude/rules/dev-phases.md. Each agent has its own domain, skills, persistent memory, and automated routines. The result is a production system that runs daily operations for a founder/CEO and supports software development workflows: from morning briefings to financial reports, community monitoring, social analytics, end-of-day consolidation, plus architectural reviews, code audits, and verified implementation pipelines.

This is not a chatbot. It is a real operating layer that runs routines, generates HTML reports, syncs meetings, triages emails, monitors community health, tracks financial metrics, and consolidates everything into a unified dashboard — all automated.

Why EvoNexus?

• Markdown-first agents — agents are .md files with system prompts, not code. No SDK, no compile step. Add an agent by dropping a file in .claude/agents/, or package reusable bundles via the plugin system (see docs/introduction.md)
• Skills as instructions — reusable capabilities are markdown too. 190+ skills covering finance, community, social, engineering, data, legal, HR, ops, product, CS
• Multi-provider by design — default runs on Anthropic's native claude CLI, but can switch to OpenRouter, OpenAI, Gemini, AWS Bedrock, Google Vertex, or Codex Auth via OpenClaude without touching a line of code. Your keys, your model choice, no vendor lock-in
• MCP integrations — first-class support for Google Calendar, Gmail, GitHub, Linear, Telegram, Canva, Notion, and more via the Model Context Protocol
• Slash commands — /clawdia, /flux, /pulse, /apex invoke agents directly from the terminal
• Persistent memory — CLAUDE.md + per-agent memory survives across sessions
• CLI-first, local-only — runs anywhere the Claude CLI (or OpenClaude) runs. Your data never leaves your infrastructure

Key Features

• Multi-Provider — runs on Anthropic (native claude) or any of 6 alternate backends via OpenClaude: OpenRouter (200+ models), OpenAI, Google Gemini, Codex Auth, AWS Bedrock, Google Vertex AI. Switch providers from the dashboard, no code changes
• 17 Core Agents + Custom — Ops, Finance, Projects, Community, Social, Strategy, Sales, Courses, Learning Retention, Personal, Knowledge, Marketing, HR, Customer Success, Legal, Product, Data — plus user-created custom-* agents (gitignored)
• 150+ Skills + Custom — organized by domain prefix (social-, fin-, int-, prod-, mkt-, gog-, obs-, discord-, pulse-, sage-, hr-, legal-, ops-, cs-, data-, pm-). Includes prod-activation-plan — the canonical skill for producing phased activation plans (index + folder-per-phase + file-per-item) used by Oracle
• 7 Core + 20 Custom Routines — daily, weekly, and monthly ADWs managed by a scheduler (core routines ship with the repo; custom routines are user-created and gitignored)
• Web Dashboard — React + Flask app with auth, roles, web terminal, service management
• 19 Integrations — Google Calendar, Gmail, Linear, GitHub, Discord, Telegram, Stripe, Omie, Bling, Asaas, Fathom, Todoist, YouTube, Instagram, LinkedIn, Evolution API, Evolution Go, Evo CRM, and more
• 2 core + custom HTML report templates — dark-themed dashboards for every domain
• Persistent Memory — two-tier system (CLAUDE.md + memory/) with LLM Wiki pattern: ingest propagation, weekly lint, centralized index, and operation log
• Knowledge Base — optional semantic search via MemPalace (local ChromaDB vectors, one-click install)
• Full Observability — JSONL logs, execution metrics, cost tracking per routine
• Heartbeats (Phase 1) — proactive agents that wake on a schedule, run a 9-step protocol, and decide whether to act
• Goal Cascade (Phase 1) — Mission → Project → Goal → Task hierarchy. Link any routine/heartbeat/ticket to a goal; agents receive the chain as prompt context
• Tickets (Phase 1) — persistent conversation/work threads with atomic checkout (UPDATE WHERE locked_at IS NULL), 6-state workflow, mention-based wake triggers, global /issues page

Screenshots

Integrations

Connect your existing tools via MCP servers, API clients, or OAuth:

Social media accounts (YouTube, Instagram, LinkedIn) are connected via OAuth through the dashboard.

Prerequisites

For the Docker install (easiest — the container ships everything):

That's it. The image includes Claude Code, Python, Node, uv, gh, and every other runtime dependency.

For the CLI / from-source install (the dev flow):

The setup wizard (make setup) checks for all prerequisites before proceeding.

AI Providers

EvoNexus runs on Anthropic's Claude by default — no extra config needed. For everything else (OpenAI, Gemini, Bedrock, OpenRouter, Vertex AI, Codex Auth), it switches to OpenClaude, a drop-in binary that speaks the Claude CLI protocol but dispatches to the provider of your choice via environment variables.