ResonanceDB 2.1.0: Context-Modulated Queries for Semantic Memory
ResonanceDB 2.1.0 introduces Context-Modulated Queries: tune per-dimension phase weights at query time for phase-aware semantic search without reindexing.
One knowledge base. Different semantic perspectives. Query-level control over relevance.
Semantic search traditionally answers a straightforward question: which information is closest to the query? But as AI systems become more sophisticated, another challenge emerges: determining which characteristics of stored knowledge matter in a particular context.
The same information can be relevant for different reasons. Scientific research may prioritize mechanisms of action, safety analysis may focus on the nature of risks, and reasoning systems may need to distinguish between different types of relationships among statements.
ResonanceDB was originally developed as a semantic memory architecture built on wave-based representations, where amplitude and phase contribute to resonance-based matching. This enables retrieval to operate on the structure of semantic signals rather than relying on a single measure of geometric proximity.
Version 2.1.0 extends this expressiveness to the query itself: applications can now control how phase structure participates in retrieval while keeping the underlying knowledge unchanged.
Context Becomes Part of the Retrieval Contract
The central innovation in ResonanceDB 2.1.0 is Parametric Phase Participation , a mechanism for controlling the contribution of individual phase dimensions.
Building on this capability, we introduce the concept of Context-Modulated Queries .
A context-modulated query provides the retrieval engine with both a query wave pattern and a configuration that determines how strongly each phase dimension contributes to relevance scoring.
Each dimension receives a weight between 0 and 1:
- 1 — full phase sensitivity;
- 0 — phase differences no longer affect the score, while amplitude remains part of the comparison;
- Intermediate values — continuous adjustment of phase sensitivity.
This gives applications the ability to select different comparison behaviors according to the task at hand.
Modulation is applied directly during resonance scoring and is also taken into account during candidate selection. It is part of query execution itself, rather than an additional filtering step applied to already retrieved results.
One Memory, Multiple Ways to Search
Consider a system working with biomedical research publications.
Its memory contains information about drugs, mechanisms of action, interactions, adverse effects, and experimental findings.
When investigating a therapeutic mechanism, the application may use one phase-sensitivity configuration. When examining safety, it may use another. When distinguishing causal relationships from associations, it may use a third.
These configurations are defined with reference to the semantic structure produced by the underlying encoder.
In every case, the system queries the same knowledge base. What changes is not the stored representation, but how its relevance to the current query is evaluated.
Switching between phase-weight configurations does not require re-encoding the corpus or building separate search indexes.
Semantic memory remains shared, while retrieval criteria become configurable parameters of interaction with that memory.
From Expressive Representations to Expressive Queries
This development advances one of the fundamental principles behind the ResonanceDB architecture.
The expressiveness of semantic memory depends on which characteristics of information it can represent and preserve. But the ability to interact with that memory also depends on how flexibly the retrieval system can use its stored structure.
ResonanceDB's wave-based model provides the foundation for storing and comparing amplitude–phase representations. Parametric phase queries introduce a second capability: controlled interpretation of those representations during retrieval.
Crucially, ResonanceDB provides more than a collection of ranking parameters. It establishes a unified software contract in which data representation, resonance comparison, and retrieval routing operate under the same phase-participation configuration.
Applications can adjust this configuration from one query to the next, obtaining reproducible results according to the selected parameters.
Expressiveness now belongs not only to stored knowledge, but also to the language used to query it.
Retrieval as a Controllable Part of Reasoning
This capability is particularly relevant to AI agents and multi-stage analytical systems.
A conventional retrieval operation accepts a query and returns relevant records. In more advanced workflows, a system can select different retrieval criteria at successive stages of a task.
For example, a research agent might:
- Retrieve materials related to a mechanism under investigation.
- Adjust its retrieval profile to examine limitations and risks.
- Query the same memory to compare supporting and conflicting evidence.
Each stage uses its own phase-participation configuration, aligned with its objective and the underlying semantic encoding model.
The same principle extends beyond biomedicine to regulatory analysis, legal document processing, research systems, and specialized RAG applications, where the relevance of retrieved information depends on the task being performed.
Context-Modulated Queries also provide a foundation for investigating retrieval sensitivity. By varying individual phase weights, developers can observe changes in ranking and examine which components of the representation influence the results.
Retrieval thus becomes a configurable instrument for working with knowledge within a broader analytical process.
An Architecture Built Around the New Contract
ResonanceDB 2.1.0 implements parametric phase participation across multiple layers of the retrieval engine.
Resonance scoring incorporates the specified weights. Phase routing and indexed retrieval adapt candidate selection to the query configuration. Weighted operations are supported by both Java implementations and the native SIMD kernel.
The release also advances adaptive candidate retrieval, improves phase routing, and strengthens the reliability of concurrent storage operations.
The new capabilities are accessible through the programming interface and REST API. Phase-participation parameters can be supplied directly with search and comparison requests.
Version 2.1.0 maintains compatibility with the 2.0.x series: existing data, storage formats, and REST endpoints remain supported, while queries without additional phase weights retain the previous scoring behavior.
This allows applications to adopt context-modulated retrieval incrementally, without modifying their existing semantic memory.
The Next Level of Interaction with Knowledge
As AI systems evolve, semantic memory must support more than storing and retrieving information. It must also provide a way to control retrieval criteria according to the task being performed.
ResonanceDB 2.1.0 makes this principle part of its architecture.
Amplitude–phase representations define the structure of stored knowledge. Context-Modulated Queries determine how that structure participates in retrieval at a given moment.
Together, they establish a model in which semantic memory and the way systems interact with it become two interconnected layers of expressiveness .
One memory. Multiple semantic perspectives. Context determines how knowledge is retrieved.