The standard

Ecosystem

STORM is neutral by design. Any model can deliberate, any source can emit a signal, any store can hold the log. This page lists what plugs in today — and what's next.

· 4 categories
· 16 integrations
· Open standard, MIT

Model providers

LLMs that act as agents in the swarm.

OpenAIlive
gpt-5, gpt-5-mini

Default coordinator. Strong structured output and tool use.

Anthropiclive
claude-sonnet, claude-haiku

Long-context reasoning for the Researcher role.

Googlelive
gemini-2.5-pro, flash

Multimodal signals; cheap parallel deliberation.

xAIbeta
grok

Realtime web grounding for the Skeptic.

Signal sources

Anything STORM can listen to and turn into a typed event.

HTTPlive
request / response

Any REST endpoint becomes a signal.

Webhookslive
inbound

Drop a URL, receive scored events.

Streamslive
SSE / WebSocket

Continuous deliberation on live feeds.

Cronlive
scheduled

Time-based agent cycles.

Storage & memory

Persistent state for replayable cycles.

Postgreslive
primary

Cycle log, agent notes, coordinator verdicts.

Redislive
cache

Hot context window for sub-second turns.

S3live
blob

Artifacts, transcripts, attachments.

pgvectorbeta
embeddings

Semantic recall across past cycles.

Runtimes

Where STORM agents actually run.

Cloudflare Workerslive
edge

Default. Cold-start-free agent execution.

Nodelive
server

Self-host the standard anywhere.

Denolive
server

Web-standard runtime, first-class TS.

Bunplanned
server

Fast local dev loop.

Bring your own everything.

STORM is a contract, not a platform. If it speaks HTTP or has a client, it joins the ecosystem.