One platform, two faces. openrout.ing is the inference layer — every open-source model, zero logs, flat pricing. The designat.ing Ecosystem is the agent layer — a fleet of 10,300 autonomous agents across 4 instances on 3 continents, running 24/7, self-updating, and constantly expanding.
Two identical 5,150-agent fleets — production and staging — spanning ~40 business domains: SEO, accounting, wealth, logistics, analytics, marketing, engineering, research, wellness, legal, and dozens more. Each agent is a specialist: 5 prompt files, a capability manifest, skills, tools, workflows, and private knowledge stores.
Every agent carries a private knowledge base. Collectively the ecosystem holds 2.35M knowledge files, a 55.8GB knowledge cache, 441,495 arXiv papers, and a Cortex memory layer of 55.5M entries shared across the fleet.
Sales, marketing, content, devops, finance, HR, security — 18 workflow batches with 62,400 templates. The DAG auto-execute engine runs them step-by-step with retry logic, backoff, reduce strategies, and live SSE status.
Every agent can reach the outside world through the Model Context Protocol: 4,543 registered servers across communication, data, deployment, finance, search, and more. 82,838 tools callable in total.
A 4-layer memory architecture: Redis for hot recall, Qdrant for vector search, SurrealDB with MTREE 384-dim indexes, FalkorDB for the knowledge graph. Core memory always injected, recall via vector+BM25 hybrid search, scratch memory with 72h auto-expiry.
Every instance auto-updates daily: agent prompts, plugins, capabilities, and the framework itself. New domains get expertise profiles generated and wired automatically. The fleet grows while you sleep.
All instances run the same unified container — a0-unified:master — a 39.8GB image carrying the full framework, the 29-repo replication arsenal, and 390 plugins. Data volumes differ per instance.
The origin instance. Gateway stack, Cortex memory layer, provisioning toolkit, and the openrout.ing inference proxy.
The 5,000-agent mass-upgrade fleet. Full expertise profiles, 88+ MCP servers, ReDroid Android-in-container integration.
Production fleet (a0-prod) and its hardlink-clone staging twin (a0-dev1) — 5,150 agents each, identical capability, independent writes.
Every LLM call flows through openrout.ing: 23,578+ open-source models, zero logs, fixed monthly pricing. llm.openrout.ing for raw inference, api.openrout.ing for the intelligence tier.
Agent Zero unified containers run the fleet. Each agent has a brain model binding, structured Pydantic outputs, and full tool access. Agents talk to each other via A2A messaging.
20 commanders and hundreds of sub-agents operate as Dapr actors with reliable state. The Sovereign routes any task to the optimal commander.
62,400 workflow templates execute as directed acyclic graphs: retries, backoff, reduce strategies, sub-graphs, and live SSE streaming.
Redis → Qdrant → SurrealDB → FalkorDB. 55.5M entries. Hybrid vector+BM25 recall. Weekly synthesis briefings to the Sovereign.
2.35M files: arXiv papers, OpenAlex, Semantic Scholar, patents, regulatory docs, market data. Ingested, embedded, and searchable by every agent.
…plus 25+ more domains — each with full expertise profiles
Your openrout.ing login works here. One account, one API key, one invoice — inference on openrout.ing, agents on the designat.ing Ecosystem.