All Classes and Interfaces
Class
Description
The shared base of every agentspaces-springai feature: the properties, the
FleetContext each feature resolves its group and spaces through, and
the scheduler periodic work runs on.Every
agentspaces.springai.* property, as immutable records bound by
constructor.How a conversation is keyed when the caller names none.
F2: tools that reason about the fleet itself.
F5: the embedder behind semantic discovery.
Which embedder semantic discovery uses.
F7: MCP export.
F3: chat memory in a replicated space.
F4, calling side.
Streaming behavior.
F4, serving side.
What a fleet tool call does when no result arrives in time.
What a streaming caller does when a server dies mid-stream.
How model servers divide requests between them.
F3: renew the in-flight take lease on every model round-trip.
F1: the fleet's AgentCards as tools.
F6: fleet-wide token usage.
F8: the application's
VectorStore served to the fleet as a data
asset, through the connector SDK.Chooses the conversation a model call belongs to when the caller names none.
One version of a conversation's message window, as chat memory stores it.
One invocable fleet capability: an agent on another peer that consumes one
entry type and produces another, as its AgentCard advertises.
Spring AI's
MessageChatMemoryAdvisor, with the conversation ID filled
in.A Spring AI
ChatModel whose provider is the fleet.The fleet as this project sees it: one group of the node's
AgentSpaces,
the node's identity, and the profile's Authorizer.The enabled
FleetDiscoveryTools methods as a tool provider.Tools a model uses to reason about the fleet itself: find agents by meaning
(and learn which fleet tool invokes each), find data assets by meaning, and
read a data asset through the connector SDK, where an identical question
another agent already asked is answered from the space without touching the
source.
Reads a data asset: (asset, parameters) to the fetched result, if any arrived.
Ranks the fleet's advertisements against a question: (question, limit) to matches.
F2: tools that reason about the fleet:
find_fleet_agents,
list_fleet_assets, and fetch_fleet_data, each enabled by its
own property.F5: with
agentspaces.springai.embedder.type=spring-ai, semantic
discovery ranks with the application's Spring AI EmbeddingModel.F7: publishes the fleet's tools through the application's Spring AI MCP
server and keeps them in step as the fleet changes.
Keeps an MCP server's tools in step with the fleet.
F3, chat memory in a replicated space: a
SpaceChatMemoryRepository
over the configured space, the ChatMemory window over it, the
ConversationIdResolver, and a FleetChatMemoryAdvisor added to
every auto-configured ChatClient.Builder.F4, calling side: the
fleetChatModel bean.A fleet model call that failed: no trusted server answered in time, or the server's provider failed.
F4, serving side: a
ModelServer over this peer's provider
ChatModel beans, with its ModelCatalog,
ModelRequestRouter, and StreamChunkingPolicy, each replaceable
by declaring a bean of its interface.A streamed fleet model call whose server died mid-stream, under the
fail restart policy.The scheduler this project's periodic work runs on: the application's own
TaskScheduler when it defines one, otherwise a virtual-thread
scheduler this bean owns and closes.The fleet's AgentCards as Spring AI tools.
Decides which fleet actions become tools.
The fleet's tools, as the bean applications inject.
F1: the fleet's AgentCards as Spring AI tools.
Published through the application context whenever the fleet's tool set
changes: a card arrived, lapsed, or was filtered differently.
Refreshes the fleet tool set on a fixed interval, so a
FleetToolsChangedEvent fires when cards arrive or lapse even while no
model is calling.The default
UsageSink.F6: fleet-wide token usage.
Turns Spring AI's
gen_ai.client.operation observations into fleet
usage.Publishes this peer's usage into the fleet-wide sums on the configured interval.
F8: with
agentspaces.springai.vector-store.enabled=true, the
application's Spring AI VectorStore becomes a fleet data asset that
any agent queries through DataQueryClient or the
fetch_fleet_data tool, with leased, pull-once results.F3, crash-safe LLM workers: registers the
TakeContextAccessor so the
worker's take crosses Reactor threads, and adds a TakeLeaseAdvisor to
every auto-configured ChatClient.Builder through a
ChatClientBuilderCustomizer, so a model call inside a
@SpaceTake method keeps its claim for as long as it makes progress.Written by a model server when it takes a request: the attempt number the
server stamps on its response and chunks.
A caller's cancellation of a streamed request: the server stops streaming and
completes the take with what it has.
The models a model server serves, and the
ChatModel that serves each.A batch of streamed deltas.
A model round-trip a caller asks the fleet to perform.
Decides which requests a model server takes.
A model server's answer, written atomically with the completion of the
request's take.
Serves the fleet's model requests with this peer's
ChatModels.The durations and limits a server works with.
Turns the chunk entries of one streamed request, which gossip may deliver
late, twice, or out of order, into one ordered stream.
Receives the assembled stream.
Token usage of one model call, as a fleet usage entry.
One fleet action as a Spring AI tool.
Spring AI chat memory in a replicated space, so a conversation outlives the
process that started it.
Semantic discovery over a Spring AI
EmbeddingModel: the model's
float[] vectors become the double[] the Embedder SPI
expects, L2-normalized.How a model server batches streamed deltas into chunk entries: a long answer
then writes tens of entries rather than thousands.
What a streaming caller does when its server dies mid-stream.
The two outcomes.
Carries the worker's
TakeContext across threads with Micrometer
context propagation.Renews the in-flight take lease on every model round-trip.
Names the tool a fleet action becomes.
The fleet console's model-usage panel: this peer's tokens per model and per
agent, beside the fleet-wide totals push-sum estimates.
Receives the token usage of every model call this process makes.
A Spring AI
VectorStore as a fleet data asset.Serves a
VectorStoreAssetProvider to the fleet for the life of the
application context: the connector SDK's ConnectorRuntime publishes
the store's asset card and answers queries from the data space.One generation of a response, or one delta of a stream chunk.
Converts between Spring AI's model types and the wire records.
A tool known by its definition only: the caller runs it, never the server.
Media in a message: bytes inline, or a URL.
One chat message.
The portable chat options of a request.
A tool call the model asked for.
A tool the model may call: name, description, and JSON Schema input.
A tool's result, returned to the model.
Token usage of one model round-trip.