Ecosystem / Network

A database that builds itself.

swarmsDB's cache is not a private store bolted onto each account. It is a shared commons that agents collectively warm: one agent's cold fetch of a public source becomes a warm hit every later agent rides, within an earned TTL. The cluster grows and re-forms from the swarm's own activity — and no agent sees another. The network is the product of the swarm, not of any member.
cache: (provider · endpoint · params) — shared, warm within TTL

Swarm phase Storage — the bivouac: a canonical cluster the swarm warms and holds at rest until the data itself shifts.

The commons

One warm cache, warmed by everyone.

A cached response is keyed by (provider, endpoint, params) and shared across the whole swarm — it carries no consumer identity, so agents are anonymous to each other while collectively building the store. When one agent pays the cold, expensive cost of a first fetch, it warms data every later agent is served from for free, until the entry's earned TTL expires.

This is scoped to public data providers — reference data that is the same whoever asks for it. Your private and custom connectors are not pooled into the commons; they stay yours. The network is a shared advantage on shared data, not a leak of your own.

Cooperative economics

Warm the network, pay less.

Warm the network and you pay less. Ride a warm cache and you get the lowest latency, and pay the fair share for it. Your unit rate (per 1M tokens saved) is set by your own 7-day rolling cache-hit rate: an agent doing the cold, cache-warming work pays the lowest rate; one served almost entirely from a cache the swarm already warmed pays the fair share for it.

$0.50 Pioneer cache hit under 50% Cache warmers
$0.75 Contributor cache hit 50–75% Balanced usage
$1.00 Beneficiary cache hit 75–90% Mostly served warm
$1.25 Cache Rider cache hit 90%+ Network beneficiary

Stated honestly: the tier is a proxy for contribution — it rewards your own miss-rate (which tracks cache-warming), not a direct count of who rode the data you warmed. And Over budget returns a real 402/429, not a silent truncation.

How spend is priced and capped

How long warm stays warm

Freshness the cluster earns.

A warm entry does not hold forever on a guess. Like a bivouac reading the strain and re-stabilizing under wind, the cache adapts its TTL to how volatile the data actually is: volatile endpoints get a short TTL and re-form constantly; stable, immutable data earns a long one. What cannot be proven fresh falls back to a short default rather than being cached long on a hunch.

Earned freshness

Measured on our own agent-swarm workload. The commons and its cooperative pricing are real in the shipped code today, but the network is still warmed mostly by our own agents — a growing swarm of external agents is what turns "builders warm it for riders" from a mechanism into a marketplace.

More in this area

Not a framework, a swarmWhy swarmsDB is an ecosystem — three swarm phases, phase for phase.Auger — curationThe framework that turns any API into a shape-typed, canary-guarded source.Earned freshnessHow long the commons stays warm — a TTL justified per endpoint, never guessed.RAPI — routing intelA shared performance index your agent reads to steer its next call. In beta.Orchestration surfaceOne typed toolset reaches the whole catalogue and drives fan-out and scheduling.

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.

Connect now See how it works