Aura v1.8.6 Patch Notes

This release adds Aura’s first real web research substrate, tightens Planner/Worker behavior, and continues cleaning out older prompt and research plumbing.

Highlights

Live Web Research

Aura now has a first-class research tool for current external information.

The Planner can call research_current_info when a user asks about current topics, recent releases, live documentation, or web search style questions. Aura performs the lookup through its own research layer and returns source/evidence data back to the model, so providers like DeepSeek can answer with current context even without native web search support.

This is the first version of Aura’s local browser-powered research path.

Playwright-Powered Browser Research

Added a new aura/research/ package with isolated modules for:

• research models
• search planning
• limits/deadlines
• browser-backed search/open behavior
• research service orchestration

The implementation uses Playwright Python with headless Chromium. In development, Chromium is installed through Playwright. In packaged builds, the build pipeline now prepares the bundled browser/runtime assets so users do not need to set up a browser manually.

One Clean Planner Tool

Research is exposed to the Planner as a single tool:

research_current_info(query, constraints)

The Worker does not get browser/search primitives directly. This keeps web research as a controlled Planner capability instead of scattering browser tools across the coding workflow.

Old Research Plumbing Removed

Cleaned out the older hidden research/Tavily-style direction and replaced it with a clearer architecture:

• one Planner-facing research tool
• isolated backend implementation
• no user API key required for search
• no Node/MCP dependency in the default research path
• fail-soft behavior when browser/search is unavailable

Prompt and Worker Cleanup