Optional AI

Start with AI off

Intake, evidence review, manual requirement comparison, direction planning and backups work without a key. Recovery AI is disabled by default. The launcher keeps AI off by default. Use --ai with an exported Anthropic key to opt in, or --no-ai to keep it off. Workspace provider settings are separate.

Set up and test the connection

read -rsp 'Anthropic API key: ' ANTHROPIC_API_KEY; echo
export ANTHROPIC_API_KEY
bash scripts/start-demo.sh --ai

Open Set up AI assistance from Start or Settings in the workspace, or from the Angular app header. Recovery currently supports Anthropic. Keep the key in the server terminal; environment files are not loaded automatically. Set a provider-side spend limit before paid reviews.

Choose Refresh AI status, then Test connection — contact Anthropic. This explicit action requests model metadata without career records or text generation. Status distinguishes unconfigured, configured but untested, connected and failed. A successful test verifies access, not advice quality. Wait 30 seconds between tests.

The integrated workspace supports setup and status. For full recovery requirement review, stop the launcher and run REBOUNDOPS_AI_ENABLED=true npm run dev with the same exported key, then open http://127.0.0.1:4200. The key stays server-side. Use REBOUNDOPS_ANTHROPIC_MODEL to override the model.

Preview and approve each review

The Angular recovery app can use server-side Anthropic review after local configuration. Follow the provider setup guide ↗ for current variables and model configuration. Environment files are not loaded automatically; keep keys server-side and out of Git.

1. Add a job target and confirm the evidence you want considered.
2. Inspect the exact JSON payload preview.
3. Consent to that request only if you want to send those contents.
4. Review suggestions yourself. They do not automatically change your session.

Claims and job requirements may contain sensitive information even when other fields are excluded. A changed session revision requires a fresh preview.

Understand limits and costs

Live recovery AI quality evaluation remains an open gate. Keep AI off for real-user pilots until that gate is met. Before a paid evaluation, set a provider-side workspace spend limit. Local request counters are not a currency ceiling and may reset without configured storage.

Workspace AI tools have separate settings and data flows. Recovery consent controls do not describe every workspace provider call. Check the tool and provider configuration before using private material.

Reference: evaluation and live-provider gate ↗. AI suggestions are advisory and do not predict hiring outcomes.