Skillget

helios-llm-orchestrator

mcp
Security Audit
Warn
Health Warn
  • License — License: MIT
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Pass
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Pass
  • Permissions — No dangerous permissions requested

No AI report is available for this listing yet.

SUMMARY

Local-first MCP multi-LLM orchestrator with task decomposition and benchmark-guided model routing through OpenRouter.

README.md

Helios

Helios lets ChatGPT or Codex run a project with a team of AI models. It splits the work into tasks, picks the best-ranked model for each one from public benchmarks, runs them through OpenRouter, checks the results, and keeps the whole project in durable memory.

How it works

  1. You describe the project, and ChatGPT or Codex breaks it into tasks.
  2. Jev, a fast decision model, sorts each task into a category such as coding, research, or vision.
  3. Helios picks the top model for that category from a weekly registry of cited public benchmarks.
  4. Tasks run through OpenRouter, or Manus for long agent jobs, and each output is checked before it is accepted.
  5. Plans, results, costs, and history are stored in SQLite, so a project survives restarts and can continue in another chat.

Quick start

You need Python 3.10+, Node.js 22+, and an OpenRouter API key.

git clone https://github.com/Erfouni/helios-llm-orchestrator.git
cd helios-llm-orchestrator
npm ci && npm test

Start Helios with your key (on Windows, see the Windows guide):

read -rs OPENROUTER_API_KEY && export OPENROUTER_API_KEY
python3 agent/server.py

Check it from another terminal:

curl http://127.0.0.1:3188/health

Then add mcp/client-config.example.json to your MCP client, with the path to your clone.

The benchmark registry starts empty. The first refresh makes one paid web search per category:

curl -X POST http://127.0.0.1:3188/benchmarks/refresh -H "Content-Type: application/json" -d '{"only_if_stale": true}'

Install as a service

  • macOS: ./scripts/install-macos.sh keeps the key in the Keychain.
  • Windows: scripts\install-windows.ps1 encrypts the key with DPAPI. See the Windows guide.
  • Linux server: use the systemd units in deploy/.

The macOS and Windows installers run the tests, start Helios at login, and refresh the benchmarks every Monday.

Tools

Area MCP tools
Run models openrouter_run_model, openrouter_compare_models, openrouter_list_models
Pick a model helios_route_task, helios_select_benchmark_model, helios_decide
Projects helios_create_project, helios_plan_project, helios_project_action, helios_run_task, helios_review_task, helios_get_project_memory
Manus manus_create_task, manus_get_task, manus_list_task_messages, manus_stop_task
Context and benchmarks helios_get_global_context, helios_update_global_context, helios_get_benchmark_registry, helios_refresh_benchmarks

Security

Helios listens only on 127.0.0.1, never returns API keys, and treats model output as untrusted data. See SECURITY.md.

Documentation

License

MIT

Reviews (0)

No results found