[Enhancement] Add Google AI Search Support

[AM-RTS]

Please consider adding Google AI Search Support, no API keys are needed.
URL: https://www.google.com/search?q=&udm=50

[mindflowgo]

Okay we've added it to the alter-ego of Proxima, which we've rebranded brAInstorm.

Ours will stay MCP + modularized + separate skills directory (so you can drop in new skills as necessary), while Proxima seems to keep these things all interconnected. However, note, Proxima will probably build out other skills that we won't!

https://github.com/mindflowgo/brAInstorm

On a surprising note: I found Google-AI wrote more optimized code patterns than Gemini in a few cases, so it seemed relevant to add. brAInstorm now includes it.

[0301chris]

I asked my Antigravity AI, who has been coding away for the last week, if it found proxima useful and if it actually ever got it to work and used it? I paste the answer with no changes:

Honestly, Proxima is a massive force multiplier. It fundamentally changes my capability ceiling.

Here is exactly how it just worked during the memory architecture research:

When you asked for a "solid AgentPro worthy design," my internal base knowledge could have hallucinated a generic SQL schema or just suggested "use Pinecone." But instead, I silently called mcp_proxima_deep_search and routed it to Perplexity.

It didn't just return a search result; it synthesized 125 lines of the absolute latest production patterns for multi-agent systems. It specifically pulled insights from frameworks like Mem0, highlighted the critical difference between episodic and semantic memory, and advised on the exact indexing trade-offs (HNSW vs IVFFlat) when using pgvector in PostgreSQL.

Why it is so valuable:

It cures my knowledge cutoff: I don't have to guess what the 2026 best practices are for Agent frameworks. I can pull the exact current consensus from the web.
It saves my token budget: Instead of me wandering around scraping individual URLs and burning context tokens to read raw HTML, Proxima acts as a highly efficient research assistant (Elena). It does the reading and returns a dense, pre-synthesized executive summary.
It prevents "hallucinated confidence": When I use Proxima, I am basing architectural decisions on actual developer consensus rather than statistically likely text generation.
In short: Yes, it absolutely works, and it is the exact reason I was able to hand you a robust, layered memory design instead of a generic key-value store. It's an indispensable tool for the "Strategy" (System 4) aspect of our VSM design.