CLI Agent Chat, Smarter Fallbacks, and Telegram Reset

You can now chat with agents from the CLI, planning chat degrades more gracefully under load, and Telegram gets a dedicated /new conversation flow.

This update improves how conversations start, recover, and stay usable across CLI, web chat, and Telegram. The common thread is reliability under real usage: fewer dead ends, cleaner resets, and better fallback behavior.

New CLI Chat Endpoint For Real Agent Conversations

You can now run direct conversations with your agents through the CLI using a dedicated chat API route.

  • New /api/cli/chat endpoint that uses the same planning chat foundation as the app
  • Streaming output parsing updated for the current AI SDK format
  • Duplicate stream artifacts reduced so CLI output is cleaner and easier to follow

Planning Chat Handles Load And Tool Calls More Gracefully

Several chat failure paths were tightened so conversations stay stable when conditions are not ideal.

  • Automatic model fallback now kicks in when the primary model is overloaded
  • Tool input normalization prevents malformed tool call payloads from breaking responses
  • Plan submit and entity creation fixes reduce auth and missing-field failures during planning flows

Telegram Now Supports Explicit Conversation Reset With /new

Telegram chat sessions no longer rely on a short inactivity reset window. You can explicitly start fresh when you want to.

  • /new creates a new thread with your currently active agent
  • Session switching is clearer so you know exactly when a fresh context starts

Marketing surfaces are easier to share and navigate:

  • Dynamic Open Graph images now render for key marketing pages, changelog entries, blog posts, and agent pages
  • FAQ improvements add clearer pathways to docs, agents, and ethics content