Changelog — NeuranAI

v3.2.4 April 21, 2026 PATCH

Console improvements & bug fixes.

Minor fixes to the Console eval runner and improvements to trace diffing UI.

FIXEDEval runner now correctly surfaces tool call errors in the trace panel instead of silently skipping them.

FIXEDSession replay no longer fails when a tool call returned an empty object.

IMPROVEDTrace diff view now highlights changed tool arguments with character-level granularity.

v3.2.0 April 8, 2026 MINOR

Shadow deployments & canary promotion controls.

You can now route a percentage of live traffic to a staged agent version and promote it to full production with one click — or roll back just as fast.

NEWShadow deployments: clone live traffic to a staged agent without affecting users. Compare resolution rates and latency side-by-side before promoting.

NEWCanary traffic slider: route 1–100% of traffic to a new agent version. Configurable per-region.

NEWOne-click rollback: revert any agent to any previous version from the Console or via API. Rollbacks complete in under 3 seconds.

IMPROVEDAgent version diff now includes a policy changelog, not just prompt diffs.

IMPROVEDSessions API now returns agent_version in every response object.

v3.1.0 March 14, 2026 MINOR

OTEL tracing, skills registry, and PT-BR improvements.

Full OpenTelemetry export for every agent trace, a browsable skills registry in Console, and significant improvements to Portuguese (Brazil) language handling.

NEWOpenTelemetry export: stream spans from every tool call and LLM call to your Grafana, Datadog, or custom OTEL collector via OTLP.

NEWSkills registry: browse, version, and reuse tools across agents from a centralized registry in the Console.

IMPROVEDPT-BR language model significantly improved for financial services vocabulary, especially collections and credit terms.

FIXEDWhatsApp channel no longer drops messages with template IDs containing hyphens.

v3.0.0 February 3, 2026 MAJOR

NeuranAI 3.0 — new runtime, new console, new evaluations.

A complete rebuild of the agent runtime and Console. Deterministic tool call execution, a new evaluation framework, and multi-region active-active failover across us-east-1 and sa-east-1.

NEWDeterministic runtime v2: every tool call is now typed with a JSON schema, executed in isolation, and included in the immutable session trace.

NEWEvaluation harness: run automated evals against a golden dataset on every deploy. Set pass/fail thresholds per metric.

NEWMulti-region active-active: sessions automatically route to the lowest-latency region and fail over within 200ms.

NEWNeuranAI Console 3.0: rebuilt dashboard with live agent monitoring, session inspector, and integrated eval runner.

DEPRECATEDv1 API endpoints (/api/v1/bots/*) deprecated. Migration guide in docs. Sunset: June 1, 2026.

v2.9.0 January 18, 2026 MINOR

Salesforce and HubSpot connectors, policy versioning.

NEWSalesforce connector: read/write to any Salesforce object with OAuth. No custom code required.

NEWHubSpot connector with contact, deal, and ticket sync.

NEWPolicies are now versioned independently of agents. A single policy change can be applied across your entire fleet.

IMPROVEDLatency p50 reduced by 18% through prompt caching on repeated system prompts.

v2.8.0 December 12, 2025 MINOR

Voice channel, Twilio integration, and ES-MX improvements.

NEWVoice channel: inbound call routing with real-time STT/TTS. Configurable voice persona per agent.

NEWTwilio connector for voice and SMS with automatic number management.

IMPROVEDES-MX model fine-tuned on 2.4M additional financial services conversations. BLEU score improved 12%.

FIXEDEmail channel now correctly handles multipart MIME messages with inline images.