Convergence

What Convergence does

Convergence combines three ideas: a Paperclip-style dashboard for companies, agents, org charts, goals, budgets and governance; a Hermes-style CEO brain with persistent memory, self-improving skills and delegation; and a Karpathy-style loop that mutates, executes, measures and keeps or reverts a change. Each company gets its own CEO agent that creates skills for what works and runs improvement loops against measurable KPIs.

The stack is React 19, Vite, Tailwind CSS 4 and shadcn/ui on the front end, a Hono API with generated OpenAPI docs and WebSocket updates, Drizzle ORM over PGlite or PostgreSQL, and BullMQ with Redis for the heartbeat scheduler and experiment runner. The CEO engine is Python (Hermes core) with a Node bridge, and it can use Ollama models such as qwen3.5 and glm-4.7-flash, the Claude API, or any OpenAI-compatible endpoint. The README links NousResearch/hermes-agent and hermes-agent-self-evolution as references.

Key features

• Dashboard for companies, agents, org charts, goals, costs and CEO insights
• Per-company CEO engine with persistent memory, skills and delegation
• Experiment loop that mutates, executes, measures and keeps or reverts changes
• Self-evolution through DSPy and GEPA optimization
• Model options include local Ollama, the Claude API and any OpenAI-compatible endpoint
• Hono API with OpenAPI docs, BullMQ queues and PGlite or PostgreSQL storage

When to use it

• Prototyping a company of agents with a CEO agent that learns from outcomes
• Running KPI-driven improvement experiments with agent-written skills
• Tracking agent budgets and goals from a single dashboard

Who it is for: Developers experimenting with multi-agent company simulations who want Hermes-style learning behind a management dashboard.

How it fits with Hermes Agent

Uses the Hermes Agent design (persistent memory and self-improving skills) for its CEO engine and references the Hermes self-evolution project, rather than being a plugin for the Hermes runtime.

FAQ