@stackbone/cli 0.9.1 → 0.9.2

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/CHANGELOG.md CHANGED
@@ -7,6 +7,26 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
7
7
 
8
8
  ## [Unreleased]
9
9
 
10
+ ## [0.9.2] - 2026-11-10
11
+
12
+ ### Added
13
+
14
+ - **A voice call starts its knowledge search before the caller finishes.**
15
+ With Flux, the voice worker, the one `stackbone dev` runs too, sends the
16
+ speech model's early guess to the box, and the box searches the agent's RAG
17
+ collection on it. A turn whose sentence matches the guess takes the passages
18
+ already found instead of searching again. Nothing is written to the
19
+ conversation and no chat model runs before the turn is real. An agent with a
20
+ model-backed input guardrail gets no early search. Each voice run records
21
+ `voiceKnowledgePrefetch` (`hit`, `miss` or `none`) in its metadata, and each
22
+ turn's metrics carry `preflightLeadMs`, how early the guess left (STA3-94).
23
+ - **The `turn.eagerThreshold` voice setting** (0.3 to 0.7, default 0.3) tunes
24
+ how early that guess comes. Deploy the worker on LiveKit Cloud before the
25
+ boxes: an older worker refuses a call whose settings carry it.
26
+ - **Model routing takes OpenRouter's whole `provider` object** on deep agents
27
+ and subagents, and the voice warm-up pins the same provider. The box applies
28
+ platform migration 0122 on boot.
29
+
10
30
  ## [0.9.1] - 2026-11-10
11
31
 
12
32
  ### Changed