Skip to content

Add opt-in Parallel Search MCP provider - #200

Draft
georgeatparallel wants to merge 1 commit into
GaiaNet-AI:mainfrom
georgeatparallel:george/parallel-search
Draft

georgeatparallel wants to merge 1 commit into
GaiaNet-AI:mainfrom
georgeatparallel:george/parallel-search

Conversation

@georgeatparallel

@georgeatparallel georgeatparallel commented Sep 25, 2026 •

Copy link
Copy Markdown

🛠️ Change Summary

This adds an opt-in Parallel Search provider to GaiaNet Node's existing Gaia Nexus chat flow. I wanted users to be able to choose an external web-search backend without turning it on by default or changing local Cardea RAG.

🔗 Related Issue

No related issue.

📋 Technical Details

Components Modified

  • Core consensus
  • P2P networking
  • RPC API
  • Storage layer
  • Cryptography
  • CLI interface
  • Configuration
  • Documentation
  • Tests

Key Changes

Added:

  • gaianet config --web-search-provider parallel and a local adapter that exposes the keyless https://search.parallel.ai/mcp endpoint to Gaia Nexus.
  • A release asset for the adapter, plus an install fallback for releases that do not include it yet.

Modified:

  • Web search stays off by default. Existing Cardea RAG and unrelated MCP entries are preserved.
  • Requests identify the project as GaiaNet-Node/0.6.0 for aggregate usage measurement. The adapter adds no user or installation identifier and does not forward authorization or cookie headers.
  • The README explains the query and context sent with search requests, along with anonymous rate limits.

Removed:

  • None.

Security Patches

No security patch is part of this PR. The local adapter keeps the upstream HTTPS host and path fixed, verifies TLS, and forwards only allowlisted protocol headers.

Benchmarks

Artificial Analysis lists Parallel Search Fast at 80.25/100 average F1 on DeepSearchQA and 19.11 seconds per task on its separate timing chart. That time combines measured search latency with derived model latency, so it is not a GaiaNet runtime measurement. Parallel Fast was last measured August 19, 2026; the page's latest-data date is September 22, 2026. DeepSearchQA results and Time per Task.

Validation

  • PYTHONDONTWRITEBYTECODE=1 python3 -m unittest discover -s tests -p 'test_parallel_search_mcp.py' -v (15 tests passed)
  • bash -n gaianet install.sh
  • python3 -m json.tool config.json
  • git diff --check

I also ran web_search and web_fetch through the pinned Gaia Nexus 0.8.2 chat caller without a Parallel key. Search returned 10 GaiaNet documentation results, and fetch returned one useful page.

Btw - I'm a devrel engineer at Parallel.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant