v1.3.0 — Commerce goes live
RunAshChat moves from a chat prototype to a working commerce surface: sellers can onboard, buyers can check out, and every payment is tracked through a database-enforced state machine.
Loading RunAsh
Releases, models, features, infrastructure, agents and breaking changes for the AI live commerce platform, newest first. Each release links to its write-up on the blog.
16 entries
RunAshChat moves from a chat prototype to a working commerce surface: sellers can onboard, buyers can check out, and every payment is tracked through a database-enforced state machine.
New sellers can go from landing page to live storefront without leaving RunAsh, and the checkout page picks up the chosen plan automatically.
A provider-agnostic payments layer, built on our own gateway rather than a third-party processor, with order status kept in sync by database triggers.
The first beta puts RunAshChat next to the stream. A studio preview, a CTA overlay editor and an agent task list let sellers run a live session in English, Hindi or Hinglish.
Agents now work alongside the stream: tasks they pick up show in a live list, and CTA timing logic decides when a call-to-action should surface.
A fine-tuned model for live-commerce chat, product Q&A and persuasive narration in English, Hindi and Hinglish, scored by an LLM-as-judge eval harness.
A control centre for live sessions, with chat and panels available in English, Hindi and Hinglish.
Hardening ahead of broader beta access.
The first post-release build replaces the prototype's browser-side API calls with a server proxy. Replies stream over SSE, can be cancelled mid-flight, and your Anthropic key never reaches the browser.
Breaking change: Set ANTHROPIC_API_KEY on the server. A key pasted into Settings remains available only as an opt-in, user-owned fallback.
All AI calls now go through /api/chat. Self-hosted deployments must provide the API key on the server.
One model string, configured in one place, so every chat and panel behaves consistently.
Groundwork that makes the chat safer and easier to extend.
The first public build of RunAshChat: an AI chat OS for live commerce, built as a prototype with simulated plugin output so the whole experience could be explored end to end.
Each persona carries its own system prompt, so the same chat can behave as a different live-commerce assistant.
A full chat workspace to explore the product before real commerce APIs were connected.
The app foundation the rest of the releases build on.
Full release write-ups live on the blog. Join the waitlist for early access to new builds.