Announcing Kanvas Ecosystem API v1.103.1: Supercharging the Agent Runtime with Neuron AI v4.1, Long-Term Memory, and Reliable Agent Wakeups

We are excited to announce the release of Kanvas Ecosystem API v1.103.1 alongside the milestone v1.103.0 release!

This release series marks one of the most substantial architectural upgrades to the Kanvas agent runtime to date. Powered by Neuron AI v4.1, Kanvas agents now feature persistent company memory, dynamic tool discovery, automated conversation compaction, durable run recovery, and a brand-new Shopping Agent—capped off by targeted reliability improvements to agent wakeup scheduling and notification delivery in v1.103.1.

---

🧠 Persistent Company Memory & Automatic Conversation Compaction

Kanvas agents are no longer bound by single-session context or forced to carry massive conversation histories into every turn.

• Cross-Conversation Company Memory: Agents can now store and recall institutional knowledge across conversations using dedicated memory search (search_memory). Interactions are indexed in the company's knowledge layer while strictly respecting agent context, company boundaries, and audience permissions.
• Automatic History Compaction: As conversations expand toward model context windows, dialogue histories are automatically summarized and archived. Agents preserve critical state, decisions, and outcomes without bloating prompt tokens or degrading inference quality.

---

⚡ Neuron AI v4.1: Dramatically Faster & More Efficient Agent Turns

Neuron AI v4.1 reimagines how agents load capabilities and process prompts. Rather than injecting every catalog tool into every prompt, agents discover tools dynamically via dynamic tool discovery (tool_search).

Internal benchmarks demonstrating the efficiency leap on identical agent questions:

By slashing token overhead and unnecessary inferences, agent response times are snappier, cost-efficient, and far more focused.

---

🛡️ Durable, Resilient, and Controllable Agent Execution

Autonomous operational processes require enterprise reliability and user control:

• Durable Run Recovery: If a background queue worker experiences job redelivery, agent runs can cleanly resume state instead of repeating completed side effects and duplicate writes.
• Turn Cancellation: Users can now stop an in-progress agent turn on demand. Kanvas validates cancellation checkpoints before running model inference or executing subsequent tools.
• Rock-Solid Wakeup Windows (v1.103.1): Refined scheduling algorithms for agent wakeup windows ensure scheduled background operations and recurring autonomous agent tasks fire predictably.

---

🛍️ Meet the New Shopping Agent: Conversational Commerce Unleashed

Expanding the roster of specialized autonomous agents (alongside Apex for AP, Arc for AR, and Expenses), Kanvas introduces the dedicated Shopping Agent.

Designed to elevate customer shopping journeys, the Shopping Agent orchestrates:

• Natural-language catalog discovery and product evaluations.
• Context-aware recommendations powered by Kanvas RAG and Typesense indexing.
• Real-time inventory and pricing checks directly within conversational messaging channels.

---

🔔 Enterprise Notification Stability (v1.103.1)

Reliable notification pipelines are vital for human-in-the-loop workflows and agent alerts. In v1.103.1, notification handling has been refactored across agent event dispatchers and operational triggers, resolving edge-case delivery failures and ensuring alerts reach Slack, email, and internal dashboards without delay.

---

🚀 Additional Enhancements

• Automotive & SalesAssist: Continuous rollout of vehicle price disclosure controls (MSRP, dealer fees, out-the-door pricing) and enriched variant-level lead intent tracking.
• Conversations & Engagement: Refined channel routing and after-hours messaging logic.
• Core Platform Stability: Updated underlying database schemas, background queue handlers, and telemetry endpoints.

---

Upgrading to v1.103.1

Kanvas Ecosystem API v1.103.1 is available now. We encourage all organizations to upgrade and take advantage of the dramatic performance improvements, persistent agent memory, and elevated runtime stability.

For full details and code-level changes, visit the Full Changelog on GitHub.