AgentSwarms-fyi /agentswarms

Unified Agentic AI and Data Platform

★ 302⑂ 66👁 4⑃ 101● 9▤ 10
TypeScriptNOASSERTIONCreated 2026-07-21agentswarms.fyi
下载:AgentSwarms 1.7.0AgentSwarms-1.7.0 ↓1.7.02026-11-14
+1
日增长
302
stars
66
forks
4
watchers
101
prs
9
issues
10
releases

README

The self-hosted agentic AI and data platform.
  Build AI agents and multi-agent swarms, land your data in a lakehouse you own,
  and put dashboards, an AI analyst and no-code machine learning on top —
  on your infrastructure, with your model keys, under one set of rules.

  
    
    
    
    
    
    
  

  
    Screenshots ·
    Features ·
    Quickstart ·
    Documentation ·
    Self-hosted vs. hosted ·
    Contributing
  


---

 AgentSwarms: agents that can act, and a data platform worth pointing them at

Most agent platforms stop at the agent and leave the data to you. AgentSwarms
ships both halves, source-available and self-hosted: agent chat, a canvas for
multi-agent swarms, RAG knowledge bases and MCP on one side; ETL pipelines, a
DuckDB-and-Parquet lakehouse you own, BI dashboards, an AI analyst that shows
its SQL, and no-code machine learning on the other. One semantic layer, one
catalog with lineage, one IAM and one hash-chained audit trail govern all of it,
so an agent asked about revenue uses the definition your finance team does, a
row filter on a table binds the dashboard and the agent alike, and every answer
traces back to the snapshot it came from. It runs from one Supabase project and
one Docker command, on your model keys.

 Why teams pick it

- One platform instead of five subscriptions — agents, RAG, ETL, a lakehouse,
  BI and ML share one login, one permission model and one audit log.
- Data you keep — tables are zstd Parquet in your own bucket with a
  transactional catalog; snapshots give time travel; nothing is locked in.
- Governance that runs before the call — model rules, spend budgets, row
  filters and column masks are enforced server-side, not reported afterwards.
- Answers you can check — the analyst cites its SQL, a forecast says what a
  period is, a trained model says when its score could mislead.
- Any model, any warehouse — OpenRouter, OpenAI, Anthropic, Gemini, Bedrock,
  Azure, OCI, Qwen, Grok, Groq, Ollama and vLLM for models; Snowflake, BigQuery,
  Databricks, Redshift, Trino, ClickHouse, Oracle, SQL Server, Postgres and MySQL
  connect directly, alongside file uploads and SaaS sources like Stripe, Shopify
  and HubSpot.
- Runs where you run — Docker on a laptop or a VM, or Kubernetes with
  autoscaling; Supabase hosted or self-hosted, so nothing has to leave your network.

 A look at it

Swarm canvas — design a multi-agent workflow as a graph and run it end to
end. Each node is a step (agent, router, condition, loop, approval, tool call);
the inspector sets its provider, model, prompt, tools and knowledge. The same
graph runs from the canvas, from the API and on a schedule — the canvas edits a
draft, and API keys and schedules keep serving the last published snapshot
until you promote it.

!The swarm canvas: a nine-node "Earnings Call Analyst Desk" workflow, with the node palette on the left and the selected agent node's configuration on the right

AI Analyst — a dedicated conversational-analysis surf

Top issues

Discussions

Latest release

AgentSwarms 1.7.0 v1.7.0 2026-11-14
AgentSwarms 1.7.0 is the first release published as images, and the first that deploys
to AWS without assembling it by hand. Every image is built for amd64 and arm64,
smoke-tested, scanned and signed. An install picks a release with AGENTSWARMS_VERSION,
and rolls back the same way. The schema now installs on a plain PostgreSQL such as
Amazon RDS. deploy/aws deploys the whole platform with Terraform: as VMs behind a load
balancer, on Amazon ECS, or on EKS from the new Helm chart. Users' code runs on Fargate
there, or on a node pool of its own. Behind a load balancer, restarts and rolling
deploy

Snapshot: 2026-11-16T09:23:14Z · Data license Apache-2.0 · JSON