runtimeai-ide

RuntimeAI check-rides from your editor — forecast eval cost, suggest scenarios, generate drafts, and run local rubrics via MCP or CLI.

No Vantage monorepo required. This package talks to the hosted RuntimeAI API at vantageai.cc.

Install

pip install runtimeai-ide

From GitHub:

pip install "git+https://github.com/simonbright/vantage.git#subdirectory=runtimeai-ide"

Editable install (Vantage repo contributors):

pip install -e ./runtimeai-ide

What works without OPENROUTER_API_KEY

Requires OPENROUTER_API_KEY + vantage-core: run / runtimeai_run_checkride — live eval with pass/fail rubric.

Cursor MCP

Add to ~/.cursor/mcp.json or your project's .cursor/mcp.json:

{
  "mcpServers": {
    "runtimeai": {
      "command": "python3",
      "args": ["-m", "runtimeai_ide", "mcp"],
      "env": {
        "RUNTIMEAI_BASE_URL": "https://www.vantageai.cc",
        "OPENROUTER_API_KEY": "${env:OPENROUTER_API_KEY}"
      }
    }
  }
}

Or run ./scripts/install-mcp.sh from this package directory.

Restart Cursor. No PYTHONPATH needed.

CLI

runtimeai-ide doctor
runtimeai-ide suggest --file models/orders.sql
runtimeai-ide forecast --scenario de_sql_optimization_v1 --turns 8
runtimeai-ide generate --file schema.sql --brief "Fix partition filter"
runtimeai-ide run --scenario de_sql_optimization_v1 --turns 1   # needs vantage-core + OPENROUTER_API_KEY

Environment

MCP tools

• runtimeai_suggest_scenario — map a file or brief to a library scenario (no key)
• runtimeai_forecast_cost — estimate per-run eval cost (no key)
• runtimeai_generate_scenario — draft a custom check-ride (no client key)
• runtimeai_run_checkride — run vantage-core locally (key + CLI required)
• runtimeai_doctor — verify API + explain key requirements

Develop

cd runtimeai-ide
pip install -e ".[dev]"
pytest -q

Publish (maintainers)

chmod +x scripts/publish-pypi.sh
export TWINE_USERNAME=__token__
export TWINE_PASSWORD=pypi-…
./scripts/publish-pypi.sh

Or tag a GitHub release — CI publishes via trusted publishing.