AgentSwarms
A self-hosted platform combining AI agents with a data stack: multi-agent workflows and RAG knowledge bases on one side, and an ETL pipeline, a lakehouse, BI dashboards and no-code machine learning on the other, governed by one shared catalog and audit trail.
AgentSwarms lets teams build AI agents and multi-agent swarms, load data into a lakehouse they own, and put dashboards, an AI analyst and no-code machine learning on top of it, all under one semantic layer and one identity and access model. A swarm canvas lets users design a multi-agent workflow as a graph of agent, router, condition, loop, approval and tool-call nodes and run it end to end.
Governance rules such as model access, spend budgets, row filters and column masks are enforced on the server before a request runs rather than only audited afterward, and a shared catalog with lineage means an agent answering a business question uses the same definitions as a dashboard. Data is stored as Parquet in a bucket the user owns, with snapshots for time travel. It connects to many model providers and warehouses, and runs from one Supabase project with a single Docker command, or on Kubernetes with autoscaling.