Releases: VoltAgent/voltagent
@voltagent/serverless-hono@2.0.12
Patch Changes
-
#1318
91abbb4Thanks @ruanchaves! - Validate workflow ownership before suspend and cancel control routes act on an execution.Fixes #1316.
-
Updated dependencies [
91abbb4]:- @voltagent/server-core@2.1.18
@voltagent/server-hono@2.0.14
@voltagent/server-elysia@2.0.9
@voltagent/server-core@2.1.18
Patch Changes
-
#1318
91abbb4Thanks @ruanchaves! - Validate workflow ownership before suspend and cancel control routes act on an execution.Fixes #1316.
-
Updated dependencies [
a1b4427]:- @voltagent/core@2.7.6
@voltagent/core@2.7.6
Patch Changes
-
#1334
a1b4427Thanks @omeraplak! - fix: prevent unhandled rejections when abortingAgent.streamText()streamsThe Problem
Agent.streamText()eagerly read the AI SDK result getters fortext,usage, andfinishReasonwhile constructing VoltAgent's wrapped result. In AI SDK v6 these fields are lazy promises, so reading them early could materialize promises that the caller never consumes.When a caller only consumed the UI/full stream and aborted the run, those unconsumed promises could reject globally as
unhandledRejectionevents.The Solution
VoltAgent now preserves the lazy getter behavior for
text,usage, andfinishReason. The sanitized text promise is also created only whenresult.textis accessed.Impact
- Aborting a consumed
streamText()stream no longer emits unhandled rejections for unconsumed result fields - Callers using only
toUIMessageStream(),toUIMessageStreamResponse(),fullStream, ortextStreamdo not need to attach defensive.catch()handlers totext,usage, orfinishReason - Matches AI SDK v6's lazy stream result contract more closely
- Aborting a consumed
@voltagent/server-core@2.1.17
Patch Changes
-
#1301
738e7b0Thanks @ruanchaves! - fix(server-core): return workflow execute result statusFixes #1300
@voltagent/core@2.7.5
Patch Changes
- #1283
23cc35aThanks @truffle-dev! - Honor the provider'sRetry-Afterheader on retried model calls. The retry loop inexecuteWithModelFallbackpreviously always used local exponential backoff capped at 10 seconds, regardless of what the server asked for; this caused concurrent agents under shared 429/503 contention to converge their retry windows. The delay now usesRetry-After(delta-seconds or HTTP-date, RFC 7231) as a floor, keeps the exponential floor as a backpressure baseline, and caps at 5 minutes for safety.
@voltagent/sandbox-blaxel@2.1.0
Minor Changes
-
#1275
74eb6f0Thanks @zrosenbauer! - Add@voltagent/sandbox-blaxel— a new workspace sandbox provider that runs your agents' shell commands inside Blaxel-managed sandboxes.import { Workspace } from "@voltagent/core"; import { BlaxelSandbox } from "@voltagent/sandbox-blaxel"; const workspace = new Workspace({ sandbox: new BlaxelSandbox({ apiKey: process.env.BL_API_KEY, workspace: process.env.BL_WORKSPACE, config: { name: "voltagent-prod", region: "us-pdx-1" }, }), });
Supports streaming stdout/stderr, per-call timeouts and
AbortSignal, output truncation, and lazy provisioning. Reach the underlying Blaxel SDK directly viasandbox.getSandbox()when you need provider-specific APIs (filesystem, previews, sessions, etc.).
@voltagent/sdk@2.0.3
Patch Changes
-
#1255
97226deThanks @omeraplak! - Add a VoltOps observability trace list API for loading persisted traces with project keys. -
Updated dependencies [
97226de,faeb4ff]:- @voltagent/core@2.7.4
@voltagent/core@2.7.4
Patch Changes
-
#1255
97226deThanks @omeraplak! - Add a VoltOps observability trace list API for loading persisted traces with project keys. -
#1257
faeb4ffThanks @omeraplak! - Fix workspace search auto-index retry behavior for tenant-aware filesystems that require operation context.