Codex Operator Center

Operator-center frontend plus a local backend that runs configurable multi-lane AI panels and streams the live reasoning thread into the UI.

Stack

• src/: React + Vite operator UI
• server/: TypeScript Node backend with SSE run streaming
• battleagents.config.json: template, provider, and judge catalog

Scripts

• npm install
• npm run dev:client
• npm run dev:server
• npm run build
• npm run server:start

Routes

• /: inbox / run launcher
• /session/:runId: live operator center for one backend run
• /session/:runId/agents: template layout sheet for the current run

Templates

Templates are defined in battleagents.config.json and displayed in the UI. The current repo ships:

• balanced
• ship-safe
• fast-answer
• web-research

Each template controls:

• lane lineup
• provider per lane
• model per lane
• judge model
• optional OpenAI web-search support

Provider tokens

Live mode reads provider credentials from .env.local if that file exists. The repo now includes a placeholder .env.local with:

• OPENAI_API_KEY
• ANTHROPIC_API_KEY
• OLLAMA_API_KEY

Backend scripts use node --env-file-if-exists=.env.local, so npm run dev:server and npm run server:start pick those values up automatically.

If a configured token is missing, the backend degrades that lane instead of crashing the run. Demo mode works without provider keys.

Running locally

1. Start the backend with npm run dev:server
2. Start the frontend with npm run dev:client
3. Open /
4. Choose a template and either demo or live
5. Start a run and open the operator center

Verification

• npm run build
• Compiled backend smoke-tested with GET /health