All Classes and Interfaces
Class
Description
Composite — drop when ALL policies agree.
Composite — drop when ANY policy says so.
Never drop.
Never drop.
Drop after at least N distinct tasks have consumed it.
Configuration for a single agent run.
An outgoing request expressed as an ordered, explicitly-labelled decomposition
of its stable and volatile parts.
Fluent assembler.
Implemented by serializers whose provider can express prompt caching
explicitly, so the placement decision survives into the wire format.
What a serializer needs to know to place cache markers, without knowing how
NARU decides cache validity.
How a provider exposes prompt caching, which determines what a caching client
must do to get a cache hit.
What a piece of model output is, independent of how it was produced.
One stable-to-volatile slice of the outgoing request context.
A tool that can be called by the agent's reasoning model.
A pending question the core needs answered before a task can continue.
How a session reaches its user.
Hash set of primitive longs using open addressing with linear probing.
Sorted set of primitive longs, ordered by value.
A single message in a chat conversation.
Abstraction over any LLM backend (Ollama, OpenAI, Anthropic, …).
One piece of session output, as produced by
NaruInteraction.write(net.thevpc.naru.api.agent.NaruLogMode, net.thevpc.nuts.text.NMsg).Response from a
NaruModelProvider chat call.The payload of a
NaruContextSegment.A run of conversation messages, in the order they must be emitted.
The tool definitions for the request.
Configures and creates one
NaruSession.A session-scoped feature that the core knows nothing about.
The on-disk catalog of saved sessions for one project.
Notified when a provider reports something the core has no opinion about: how many
tokens a call cost, and what the provider's own rate limits currently are.
One normalised piece of model output.
The one
NaruStreamChunk implementation.A
NaruStreamHandler that keeps what it is given.Local stand-in for
BooleanSupplier, to avoid a JDK-vs-android signature debate.Receives model output as it is produced.
Turns a model's output into the one form everything downstream understands:
an ordered list of
NaruStreamChunks, and an assistant message whose
reasoning is already separated from its answer.Where a piece of thinking content came from.
One continuous stretch of model reasoning, kept separate from the visible
answer on
NaruMessage.Splits a text stream into thinking and answer content using a delimiter pair.
The delimiters a model wraps its reasoning in, when it puts reasoning in the
answer text instead of a field of its own.
Which part of a model call a token belongs to.
A tool that can be called by the agent's reasoning model.
A tool call requested by the model inside an assistant message.
OpenAI-compatible tool definition (Ollama uses the same format).
Builder for a single parameter inside a JSON schema "properties" block.
Drop after exactly one consumption — unicast semantics.
Drop after a fixed duration since firedAt.