Skip to content

v2026.03.09

Choose a tag to compare

@elophanto elophanto released this 09 Mar 13:25
· 434 commits to main since this release

Highlights

VS Code Extension

IDE-integrated chat sidebar that connects to the EloPhanto gateway as another channel adapter. Chat with the agent from VS Code with full IDE context injection — active file, selection, diagnostics, open files. Same conversation across all channels.

  • Chat panel with streaming responses, crop marks, tool step indicators
  • Chat history panel — load and switch between past conversations
  • Tool approvals via native VS Code notifications with risk classification
  • Right-click context menu: Send Selection, Explain This Code, Fix This Code
  • Status bar with connection state (connected/disconnected/thinking)
  • Gateway checker on startup (does not auto-launch — vault password requires terminal)
  • New Chat button (UI-only clear, never wipes agent memory)
  • Close panel button, tooltips on all controls
  • Design matches the web dashboard aesthetic

Business Launcher Skill

New business-launcher skill — 7-phase pipeline to spin up revenue-generating businesses end-to-end.

  • All business types: SaaS, local service, professional service, ecommerce, digital product, content/media
  • B2B vs B2C classification drives pricing, content style, launch channels, growth tactics
  • 7 phases: Classify → Validate → Plan → Build → Deploy → Launch → Grow → Operate
  • Smart payment handling: checks knowledge_search, vault_lookup, wallet_status for existing credentials before asking the owner
  • Cross-session execution: goal checkpoints, knowledge persistence, autonomous mind continuation
  • Type-specific launch channels: tech → Product Hunt/HN; local → Google Business/Yelp/Nextdoor; B2B → LinkedIn/email outreach

OpenAI as First-Class Provider

  • GPT-5 and OpenAI models fully supported alongside Claude
  • Smart tool prioritization for OpenAI's 128-tool limit
  • Tool profiles for context-aware filtering
  • Temperature parameter fix for GPT-5 models

Autonomous Experimentation

New experimentation tools inspired by karpathy/autoresearch — the agent can design, run, and analyze experiments autonomously.

All Changes

New Features

  • VS Code extension with IDE chat sidebar, tool approvals, context injection
  • Business launcher skill (7-phase pipeline, all business types, B2B/B2C)
  • OpenAI as first-class LLM provider (GPT-5 support)
  • Autonomous experimentation tools
  • Tool profiles for context-aware tool filtering
  • Smart tool prioritization for 128-tool limit
  • Project-naming skill with auto-injection
  • Cloud deployment infrastructure (Fly.io hosted SaaS)
  • Cross-session search, code execution sandbox
  • Enhanced skill security and proactive nudging
  • Expanded init wizard (14 sections)

Bug Fixes

  • Fix identity beliefs stored as string instead of dict
  • Fix tool trimming dropping actively-used tools mid-conversation
  • Fix ToolResult.to_dict crash when data is not a dict
  • Fix autonomous mind state lost on restart
  • Fix GPT-5 models rejecting temperature parameter
  • Fix mypy errors in experimentation module
  • Strip Playwright automation defaults for real browser stealth
  • Fix GitHub URL in pyproject.toml

Docs

  • 43 documentation pages (up from 41)
  • New: VS Code Extension (doc 43), Business Launcher (doc 42)
  • Updated README (EN + CN) with VS Code, business launcher
  • Updated CHANGELOG, docs/README

Full Changelog: v2026.03.06...v2026.03.09