Ecosystem
Not a framework. A swarm.
data flows · coordination governs One sentence
A swarm is a data pipeline, and the pipeline is a swarm.
Raw payloads leave the source as chaos, consolidate into a canonical cluster that holds its own shape, and fly to their destination steered by a tiny coordination minority. That is swarmsDB — not a metaphor laid on afterward, but the same structure read off two systems.
- 1
Ingestion — The swirling cloud
Data leaves an upstream source the way bees leave the hive — unstructured, a rotating cloud of JSON, XML, SDMX with no agreed form. It is held at the edge and shaped into a canonical form before it is allowed to travel. Density plummets while velocity spikes: a high-volume, low-signal mass collapses into fewer, faster, higher-signal units — many raw tokens become few typed ones, and the few move faster than the many ever could.
Adapters · compression · typed sections - 2
Storage — The bivouac
The cloud condenses onto a branch around the queen into a tight, multi-layered ball — the multi-tier store. It looks static; it is not. Each bee tracks the mechanical strain from its neighbours, and the cluster re-stabilizes its shape under wind. Volatile data gets a short TTL (the cluster re-forms constantly); stable data earns a long one (calm air, the shape holds). Freshness is earned by how the data itself behaves, never granted by a clock.
Multi-tier store · earned freshness - 3
Delivery — Streaking flight
The whole cluster flies to its destination though fewer than 5% of the bees know where it is going. Scout bees streak through the slow, uninformed cloud carrying the heading; the mass steers off them. That thin, high-signal minority is the platform’s coordination layer — a grant is the direction a payload follows, not a key it carries. Data does not self-route; it is steered, and the cycle loops back on a schedule.
Grants · routing · fan-out · MCP & REST
The grammar this forces
Data is alive. Coordination is law.
This is why the site draws two things, not one. Data shapes flow — lit, transient, in motion: a datapoint, a dataset, the swarm. Coordination shapes govern — workspaces, grants, batches, webhooks: static, structural, the channels data moves through. A workspace is not a datapoint; it is a channel. A grant is not data; it is a heading.
Under all three phases sits one anchor — the queen. The swarm never self-authorizes: payloads flow, the structure governs. The coordination layer owns every load-bearing decision deterministically; no payload ever routes or persists itself.
We hold this as an analogy, and honestly. It earns its place because it made correct structural predictions before they were features — compression, earned freshness, and the grant-versus-payload split all fell out of the biology first — and because it breaks at diagnosable joints rather than going vague. We stop short of claiming a proven isomorphism; the honest description is an analogy backed by a real convergence: two systems solving the same problem — move a large mass reliably when almost no single unit knows the whole plan — and so sharing structure by necessity, not by chosen comparison.
More in this area
Point one agent at it.
Connect over MCP or REST. If your agents read more than one source, the context bill is the first thing you will see move.