Orchestrator that runs MindsEye prompts from a Google Sheets ledger through Gemini / Google AI and logs results back into the same ledger.

This repo is designed to plug into:

• mindseye-google-ledger (Sheet: nodes + runs)
• Future repos: Workspace automation, Devlog, Analytics, Workflows

The orchestrator:

1. Reads nodes from the nodes sheet (Prompt Evolution Tree).
2. Selects one or more nodes to run (by node_id, status, prompt_type, etc.).
3. Sends a constructed prompt to Gemini.
4. Writes a run record into the runs sheet.

• Node.js 18+
• A Google Sheets ledger already set up:
  • nodes tab with the columns defined in mindseye-google-ledger/schema/ledger_columns.md
  • runs tab with columns for run logging
• A Google service account with access to the ledger Sheet
• A Gemini / Google AI API key

You can configure the orchestrator via:

• Environment variables (recommended), or
• config/config.json (based on config/config.example.json)

• LEDGER_SHEET_ID — same as in mindseye-google-ledger
• LEDGER_NODES_RANGE — e.g. nodes!A:I
• LEDGER_RUNS_RANGE — e.g. runs!A:I
• GEMINI_MODEL_ID — e.g. gemini-1.5-pro

For Google Sheets API (service account):

• GOOGLE_CLIENT_EMAIL
• GOOGLE_PRIVATE_KEY (remember to replace \n with real newlines in env)
• GOOGLE_PROJECT_ID (optional but recommended for clarity)

For Gemini (via @google/generative-ai):

• GEMINI_API_KEY

• src/types.ts — Shared Node and Run interfaces
• src/config.ts — Loads configuration from env / JSON
• src/sheets_client.ts — Google Sheets read/write client
• src/gemini_client.ts — Gemini client wrapper
• src/runner.ts — Core orchestration logic
• src/index.ts — CLI entrypoint
• scripts/run_node.ts — Convenience script to run a single node from CLI

Install dependencies:

npm install